Spaces:
Running on Zero
Running on Zero
Upload 43 files
Browse files- .gitattributes +4 -0
- DEPLOY.md +11 -0
- README.md +30 -22
- app.py +335 -77
- assets/anomaly_confusion_matrix.png +0 -0
- assets/anomaly_metrics_bar.png +0 -0
- assets/anomaly_roc_curve.png +0 -0
- assets/intent_confusion_matrix.png +0 -0
- assets/intent_dataset_composition.png +0 -0
- assets/intent_per_class_bar.png +0 -0
- assets/latency_bar.png +0 -0
- assets/retrieval_hitrate_bar.png +0 -0
- assets/sensor_distributions.png +0 -0
- assets/vision_metrics_bar.png +0 -0
- assets/vision_noise_robustness.png +0 -0
- assets/vision_sample_annotated.png +3 -0
- assets/vision_sample_raw.png +3 -0
- build_artifacts.py +73 -0
- data/intent_dataset.csv +135 -135
- data/latency_eval.json +3 -3
- data/retrieval_eval.json +68 -12
- data/sample_sensor_batch.csv +41 -0
- data/vision_eval.json +24 -0
- models/anomaly_iforest.joblib +1 -1
- models/anomaly_scaler.joblib +1 -1
- models/intent_pipeline.joblib +2 -2
- requirements.txt +3 -0
- src/alerting.py +62 -0
- src/data_generation.py +7 -0
- src/knowledge_base.py +55 -0
- src/llm_client.py +28 -4
- src/vision_model.py +164 -0
.gitattributes
CHANGED
|
@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
assets/intent_confusion_matrix.png filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
assets/sensor_distributions.png filter=lfs diff=lfs merge=lfs -text
|
| 38 |
+
assets/vision_sample_annotated.png filter=lfs diff=lfs merge=lfs -text
|
| 39 |
+
assets/vision_sample_raw.png filter=lfs diff=lfs merge=lfs -text
|
DEPLOY.md
CHANGED
|
@@ -117,3 +117,14 @@ reproducible ML work rather than just a UI demo. Since everything here is
|
|
| 117 |
built on synthetic, self-generated data, this same Space link works as-is
|
| 118 |
for applications to multiple companies — no company-specific data or
|
| 119 |
branding needs to change.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
built on synthetic, self-generated data, this same Space link works as-is
|
| 118 |
for applications to multiple companies — no company-specific data or
|
| 119 |
branding needs to change.
|
| 120 |
+
|
| 121 |
+
## Optional secrets
|
| 122 |
+
|
| 123 |
+
| Secret | Purpose |
|
| 124 |
+
|---|---|
|
| 125 |
+
| `HF_TOKEN` (or `Smart_Warehouse`, see `src/llm_client.py`) | Enables LLM-generated answers (otherwise retrieval-only fallback) |
|
| 126 |
+
| `SLACK_WEBHOOK_URL` | Makes the Predictive Maintenance "Send Slack alert" option actually post to Slack (otherwise it shows a labelled preview) |
|
| 127 |
+
|
| 128 |
+
**Dependency note:** `requirements.txt` pins `pydantic==2.10.6`. Gradio 5.9.1's API-schema generation
|
| 129 |
+
crashes on newer pydantic releases (`TypeError: argument of type 'bool' is not iterable`), which would
|
| 130 |
+
make the Space fail to start after a rebuild. Don't remove the pin unless you also upgrade Gradio.
|
README.md
CHANGED
|
@@ -8,8 +8,9 @@ sdk_version: 5.9.1
|
|
| 8 |
app_file: app.py
|
| 9 |
pinned: false
|
| 10 |
license: mit
|
| 11 |
-
short_description: LLM warehouse copilot
|
| 12 |
---
|
|
|
|
| 13 |
# 🏭 Smart Warehouse AI Assistant
|
| 14 |
|
| 15 |
An LLM-powered, retrieval-grounded AI copilot for automated warehouse
|
|
@@ -24,26 +25,21 @@ metrics on held-out test data.
|
|
| 24 |
|
| 25 |
## Tabs
|
| 26 |
|
| 27 |
-
1. **💬 AI Assistant** —
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
ROC-AUC, retrieval hit-rate, and latency benchmarks, all computed on
|
| 43 |
-
held-out data by `build_artifacts.py` and rendered as charts (bar
|
| 44 |
-
charts, confusion matrices, ROC curve, feature-distribution histograms)
|
| 45 |
-
alongside the underlying tables.
|
| 46 |
-
5. **ℹ️ About** — project write-up, architecture diagram, tech stack.
|
| 47 |
|
| 48 |
## Quick start (local)
|
| 49 |
|
|
@@ -89,6 +85,8 @@ smart-warehouse-ai/
|
|
| 89 |
├── build_artifacts.py # generates data, trains models, evaluates, saves plots
|
| 90 |
├── requirements.txt
|
| 91 |
├── src/
|
|
|
|
|
|
|
| 92 |
│ ├── data_generation.py # synthetic intent / inventory / sensor datasets
|
| 93 |
│ ├── knowledge_base.py # warehouse-ops knowledge base (RAG source docs)
|
| 94 |
│ ├── retriever.py # TF-IDF retriever
|
|
@@ -115,7 +113,8 @@ benchmark). Headline numbers from the included run:
|
|
| 115 |
| Anomaly detector | Recall | ~97% |
|
| 116 |
| Anomaly detector | F1 | ~98% |
|
| 117 |
| Anomaly detector | ROC-AUC | ~1.00 |
|
| 118 |
-
| RAG retriever | Hit-rate@2 | 100% |
|
|
|
|
| 119 |
|
| 120 |
*(Computed on synthetic, held-out test data — see the Evaluation tab for
|
| 121 |
methodology notes.)*
|
|
@@ -150,3 +149,12 @@ is about honestly picking the cutoff, not about detection ability.
|
|
| 150 |
## License
|
| 151 |
|
| 152 |
MIT — feel free to fork and adapt.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
app_file: app.py
|
| 9 |
pinned: false
|
| 10 |
license: mit
|
| 11 |
+
short_description: LLM warehouse copilot, predictive maintenance & CV
|
| 12 |
---
|
| 13 |
+
|
| 14 |
# 🏭 Smart Warehouse AI Assistant
|
| 15 |
|
| 16 |
An LLM-powered, retrieval-grounded AI copilot for automated warehouse
|
|
|
|
| 25 |
|
| 26 |
## Tabs
|
| 27 |
|
| 28 |
+
1. **💬 AI Assistant** — free-text warehouse-ops Q&A. Intent classification → TF-IDF retrieval over a
|
| 29 |
+
17-article knowledge base (RAG) → hosted LLM, with **multi-turn memory** (follow-ups like
|
| 30 |
+
*"what about Zone D instead?"* work), a multi-model fallback chain, a retrieval-only fallback, and a
|
| 31 |
+
"Test LLM connection" diagnostics button.
|
| 32 |
+
2. **📦 Inventory & Order Query** — natural-language queries over synthetic inventory / order tables.
|
| 33 |
+
3. **⚠️ Predictive Maintenance** — Isolation Forest anomaly detector (calibrated threshold). Single
|
| 34 |
+
readings via sliders, **CSV batch scoring** (sample file provided), and optional **Slack alerts**
|
| 35 |
+
(real if `SLACK_WEBHOOK_URL` is set, otherwise a clearly-labelled preview — never claimed as sent).
|
| 36 |
+
4. **📷 Visual Inventory Scan** — a *classical* (non-deep-learning) OpenCV pipeline that counts boxes and
|
| 37 |
+
estimates rack occupancy on synthetic rack images, with an adjustable noise slider.
|
| 38 |
+
5. **📊 Model Evaluation** — graphical, held-out metrics for every component, including the vision
|
| 39 |
+
pipeline's accuracy-vs-noise curve.
|
| 40 |
+
6. **ℹ️ About** — dataset provenance, model function table, architecture, how to test.
|
| 41 |
+
|
| 42 |
+
A **🌐 Language / 言語** selector switches the explanatory text between English and 日本語.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
## Quick start (local)
|
| 45 |
|
|
|
|
| 85 |
├── build_artifacts.py # generates data, trains models, evaluates, saves plots
|
| 86 |
├── requirements.txt
|
| 87 |
├── src/
|
| 88 |
+
│ ├── vision_model.py # classical-CV rack scanner (+ synthetic scene generator)
|
| 89 |
+
│ ├── alerting.py # Slack alert (real webhook or preview) + email preview
|
| 90 |
│ ├── data_generation.py # synthetic intent / inventory / sensor datasets
|
| 91 |
│ ├── knowledge_base.py # warehouse-ops knowledge base (RAG source docs)
|
| 92 |
│ ├── retriever.py # TF-IDF retriever
|
|
|
|
| 113 |
| Anomaly detector | Recall | ~97% |
|
| 114 |
| Anomaly detector | F1 | ~98% |
|
| 115 |
| Anomaly detector | ROC-AUC | ~1.00 |
|
| 116 |
+
| RAG retriever | Hit-rate@1 / @2 | 88% / 100% |
|
| 117 |
+
| Vision (box counting, noise 0.4) | Exact count match | ~88% (100% clean → ~77% at max noise) |
|
| 118 |
|
| 119 |
*(Computed on synthetic, held-out test data — see the Evaluation tab for
|
| 120 |
methodology notes.)*
|
|
|
|
| 149 |
## License
|
| 150 |
|
| 151 |
MIT — feel free to fork and adapt.
|
| 152 |
+
|
| 153 |
+
## Vision pipeline — honest evaluation notes
|
| 154 |
+
|
| 155 |
+
The first detector (Canny edges + dilation) merged every box into one blob; it was replaced with
|
| 156 |
+
colour-distance segmentation. On clean renders it then hit a suspicious 100%, so a `noise_level`
|
| 157 |
+
(pixel noise, brightness/contrast jitter, per-box colour variation) was added and the headline number is
|
| 158 |
+
reported at a moderate noise level alongside a full noise-sweep chart. Occupancy was also redefined from
|
| 159 |
+
padded pixel area to *occupied grid cells* to match how ground truth is defined. All scenes are synthetic —
|
| 160 |
+
a real deployment would fine-tune a detector (e.g. YOLO) on real photos.
|
app.py
CHANGED
|
@@ -27,6 +27,8 @@ import pandas as pd
|
|
| 27 |
from src.anomaly_model import FEATURES as SENSOR_FEATURES
|
| 28 |
from src.anomaly_model import load_artifacts as load_anomaly_artifacts
|
| 29 |
from src.anomaly_model import score_reading
|
|
|
|
|
|
|
| 30 |
from src.data_generation import generate_inventory_db, generate_orders_db
|
| 31 |
from src.intent_model import INTENT_DESCRIPTIONS, load_pipeline, predict as intent_predict
|
| 32 |
from src.inventory_db import query_inventory, query_orders
|
|
@@ -104,6 +106,7 @@ intent_eval = _load_json("intent_eval.json")
|
|
| 104 |
anomaly_eval = _load_json("anomaly_eval.json")
|
| 105 |
retrieval_eval = _load_json("retrieval_eval.json")
|
| 106 |
latency_eval = _load_json("latency_eval.json")
|
|
|
|
| 107 |
|
| 108 |
HF_TOKEN_SET = bool(_get_hf_token())
|
| 109 |
|
|
@@ -136,7 +139,9 @@ except ImportError:
|
|
| 136 |
|
| 137 |
def chat_fn(message, history):
|
| 138 |
intent = intent_predict(intent_pipeline, message)
|
| 139 |
-
|
|
|
|
|
|
|
| 140 |
|
| 141 |
intent_label = INTENT_DESCRIPTIONS.get(intent.intent, intent.intent)
|
| 142 |
meta_lines = [f"**Detected intent:** {intent_label} ({intent.confidence:.0%} confidence)"]
|
|
@@ -215,20 +220,17 @@ def inventory_query_fn(text):
|
|
| 215 |
# TAB 3 -- Predictive Maintenance / Anomaly Detection
|
| 216 |
# ==========================================================================
|
| 217 |
|
| 218 |
-
def anomaly_fn(motor_temp, vibration, current, belt_speed):
|
| 219 |
reading = {
|
| 220 |
-
"motor_temp_c": motor_temp,
|
| 221 |
-
"
|
| 222 |
-
"current_amps": current,
|
| 223 |
-
"belt_speed_mps": belt_speed,
|
| 224 |
}
|
| 225 |
result = score_reading(anomaly_model, anomaly_scaler, reading, threshold=anomaly_threshold)
|
| 226 |
verdict = "🔴 ANOMALY DETECTED" if result.is_anomaly else "🟢 Normal operating range"
|
| 227 |
detail = (
|
| 228 |
f"### {verdict}\n\n"
|
| 229 |
f"**Anomaly score:** {result.anomaly_score:.2f} / 1.00\n\n"
|
| 230 |
-
f"| Sensor | Value | Typical normal range |\n"
|
| 231 |
-
f"|---|---|---|\n"
|
| 232 |
f"| Motor temperature | {motor_temp:.1f} °C | 47–63 °C |\n"
|
| 233 |
f"| Vibration | {vibration:.2f} mm/s | 1.2–3.2 mm/s |\n"
|
| 234 |
f"| Motor current | {current:.1f} A | 9–15 A |\n"
|
|
@@ -241,6 +243,14 @@ def anomaly_fn(motor_temp, vibration, current, belt_speed):
|
|
| 241 |
"wear, belt misalignment, or motor overload -- schedule maintenance before "
|
| 242 |
"the next shift to avoid an unplanned stoppage."
|
| 243 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 244 |
return detail
|
| 245 |
|
| 246 |
|
|
@@ -255,6 +265,60 @@ def load_preset(name):
|
|
| 255 |
return ANOMALY_PRESETS[name]
|
| 256 |
|
| 257 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 258 |
# ==========================================================================
|
| 259 |
# TAB 4 -- Model Evaluation
|
| 260 |
# ==========================================================================
|
|
@@ -327,7 +391,11 @@ Modern warehouse automation platforms — automated storage & retrieval
|
|
| 327 |
systems (AS/RS), conveyor & sortation lines, AGVs/AMRs, and the WMS/WCS
|
| 328 |
software that orchestrates them — generate huge volumes of operational
|
| 329 |
data: equipment telemetry, transactions, safety logs, and ad-hoc questions
|
| 330 |
-
from floor staff.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 331 |
|
| 332 |
## What this demonstrates
|
| 333 |
|
|
@@ -350,12 +418,12 @@ being realistic enough to demonstrate the underlying ML techniques properly.
|
|
| 350 |
|
| 351 |
| Dataset | What it is | Size | How it's generated |
|
| 352 |
|---|---|---|---|
|
| 353 |
-
| **Knowledge base** | Original, hand-written articles on generic warehouse-automation concepts (AS/RS, AGV/AMR, WMS, sortation, picking strategy, predictive maintenance, safety, inventory accuracy, KPIs, energy efficiency) |
|
| 354 |
| **Intent queries** | Example free-text operational questions/requests across 8 categories | ~480 examples | ~8 templates per category with randomised SKU codes, zone names, order IDs, and equipment IDs slotted in |
|
| 355 |
| **Inventory table** | SKU records with category, zone, on-hand units, reorder point, unit cost | 60 SKUs | Randomised within realistic ranges (fixed seed) |
|
| 356 |
| **Orders table** | Order records with status, line count, priority, zone | 80 orders | Randomised within realistic ranges (fixed seed) |
|
| 357 |
| **Sensor readings** | Conveyor/crane motor telemetry: temperature, vibration, current, belt speed | 1,000 readings (900 normal + 100 anomalous) | Normal readings drawn from realistic operating ranges; anomalies simulate known failure signatures (elevated temp/vibration/current + reduced belt speed) |
|
| 358 |
-
| **Retrieval eval set** | Hand-labelled (question → expected knowledge-base article) pairs |
|
| 359 |
|
| 360 |
## How each model works (function)
|
| 361 |
|
|
@@ -441,11 +509,19 @@ chosen to be as simple as it can be while still doing that job well and
|
|
| 441 |
being honestly evaluated, rather than reaching for the biggest available
|
| 442 |
model by default.
|
| 443 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 444 |
## Limitations & next steps
|
| 445 |
|
| 446 |
- All data here is **synthetic**, for portfolio/demo purposes — a production
|
| 447 |
version would connect to real WMS/WCS APIs and historical sensor logs.
|
| 448 |
-
- The intent set (8 classes) and knowledge base (
|
| 449 |
small to keep the demo fast and auditable; both are easy to extend.
|
| 450 |
- The anomaly detector uses 4 hand-picked features; a production system would
|
| 451 |
likely use a richer multivariate sensor set and a supervised or
|
|
@@ -476,15 +552,175 @@ CUSTOM_CSS = """
|
|
| 476 |
.data-badge b { color: #1e3a8a; }
|
| 477 |
"""
|
| 478 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 479 |
def data_badge(html: str) -> str:
|
| 480 |
return f'<div class="data-badge">{html}</div>'
|
| 481 |
|
| 482 |
with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="Smart Warehouse AI Assistant") as demo:
|
| 483 |
-
gr.Markdown(
|
| 484 |
-
|
| 485 |
-
|
| 486 |
-
"intent routing · predictive maintenance · retrieval-grounded Q&A</p>"
|
| 487 |
-
)
|
| 488 |
if not HF_TOKEN_SET:
|
| 489 |
gr.Markdown(
|
| 490 |
"> ⚠️ **No HF token secret detected.** The AI Assistant tab will run in "
|
|
@@ -494,22 +730,8 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="
|
|
| 494 |
)
|
| 495 |
|
| 496 |
with gr.Tab("💬 AI Assistant"):
|
| 497 |
-
with gr.Accordion("
|
| 498 |
-
gr.HTML(data_badge(
|
| 499 |
-
"🧪 <b>Source:</b> hand-written by the developer — 10 original knowledge-base "
|
| 500 |
-
"articles on warehouse operations (AS/RS, AGV/AMR, WMS, sortation, picking, "
|
| 501 |
-
"maintenance, safety) plus ~480 template-generated example queries. "
|
| 502 |
-
"Not scraped or sourced from any real company.<br>"
|
| 503 |
-
|
| 504 |
-
"<b>What it describes:</b> generic intralogistics/automation concepts and "
|
| 505 |
-
"operational scenarios (equipment faults, safety incidents, order/inventory "
|
| 506 |
-
"questions) — general domain knowledge, not any specific facility's live data.<br>"
|
| 507 |
-
|
| 508 |
-
"<b>Function:</b> answers free-text warehouse-ops questions by first classifying "
|
| 509 |
-
"the query's intent, retrieving the most relevant knowledge-base passage(s) "
|
| 510 |
-
"(RAG), then generating a grounded answer with a hosted LLM — falling back to "
|
| 511 |
-
"showing the retrieved passages directly if no LLM is available."
|
| 512 |
-
))
|
| 513 |
gr.ChatInterface(
|
| 514 |
fn=chat_fn,
|
| 515 |
type="messages",
|
|
@@ -528,27 +750,10 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="
|
|
| 528 |
test_llm_btn.click(test_llm_fn, inputs=None, outputs=test_llm_output)
|
| 529 |
|
| 530 |
with gr.Tab("📦 Inventory & Order Query"):
|
| 531 |
-
with gr.Accordion("
|
| 532 |
-
gr.HTML(data_badge(
|
| 533 |
-
f"🧪 <b>Source:</b> randomly generated by the developer's code "
|
| 534 |
-
f"(<code>src/data_generation.py</code>, fixed seed) — "
|
| 535 |
-
f"{len(inventory_df)} synthetic SKU records and "
|
| 536 |
-
f"{len(orders_df)} synthetic orders. "
|
| 537 |
-
"Not exported from any real company's WMS.<br>"
|
| 538 |
-
|
| 539 |
-
"<b>What it describes:</b> a stand-in inventory table "
|
| 540 |
-
"(SKU, category, zone, on-hand units, reorder point, unit cost) "
|
| 541 |
-
"and orders table (order ID, status, line count, priority, zone) — "
|
| 542 |
-
"realistic in shape and ranges, but fictional records, not live "
|
| 543 |
-
"warehouse data.<br>"
|
| 544 |
-
|
| 545 |
-
"<b>Function:</b> classifies whether a query is about inventory or "
|
| 546 |
-
"order status, extracts SKU/order-ID/zone identifiers with regex, "
|
| 547 |
-
"and filters the corresponding table — a lightweight stand-in "
|
| 548 |
-
"for a natural-language WMS query tool."
|
| 549 |
-
))
|
| 550 |
with gr.Row():
|
| 551 |
-
inv_input = gr.Textbox(label="Query", placeholder="How many units of SKU-1042 are in Zone
|
| 552 |
inv_btn = gr.Button("Search", variant="primary", scale=1)
|
| 553 |
inv_note = gr.Markdown()
|
| 554 |
inv_result = gr.Dataframe(label="Results", wrap=True)
|
|
@@ -557,11 +762,11 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="
|
|
| 557 |
|
| 558 |
gr.Examples(
|
| 559 |
examples=[
|
| 560 |
-
"How many units of SKU-1042 are in Zone
|
| 561 |
"What's the status of order #10007?",
|
| 562 |
"Show me low stock items",
|
| 563 |
"Any delayed orders?",
|
| 564 |
-
"Is SKU-1015 in stock at Zone
|
| 565 |
"Do we have enough SKU-1030 to fulfill 200 units?",
|
| 566 |
"Track order #10021 for me",
|
| 567 |
"Show the fulfillment status of #10045",
|
|
@@ -575,20 +780,8 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="
|
|
| 575 |
gr.Dataframe(orders_df, wrap=True)
|
| 576 |
|
| 577 |
with gr.Tab("⚠️ Predictive Maintenance"):
|
| 578 |
-
with gr.Accordion("
|
| 579 |
-
gr.HTML(data_badge(
|
| 580 |
-
"🧪 <b>Source:</b> randomly generated by the developer's code "
|
| 581 |
-
"(`src/data_generation.py`, fixed seed) — 1,000 synthetic sensor readings "
|
| 582 |
-
"(900 normal + 100 simulated-fault). Not logged from real equipment.<br>"
|
| 583 |
-
"<b>What it describes:</b> conveyor/crane motor sensor readings — temperature, "
|
| 584 |
-
"vibration, current draw, and belt speed — with the 'anomaly' readings modelled "
|
| 585 |
-
"on realistic failure signatures (elevated temp/vibration/current with reduced "
|
| 586 |
-
"belt speed, as seen with bearing wear or motor overload).<br>"
|
| 587 |
-
"<b>Function:</b> an Isolation Forest model trained unsupervised (it never sees "
|
| 588 |
-
"an anomaly label) learns what 'normal' sensor behaviour looks like, then flags "
|
| 589 |
-
"any new reading — like the ones you enter below — that deviates from it, as an "
|
| 590 |
-
"early-warning signal before an unplanned equipment stoppage."
|
| 591 |
-
))
|
| 592 |
preset_dropdown = gr.Dropdown(
|
| 593 |
choices=list(ANOMALY_PRESETS.keys()), label="Load a preset reading", value="Normal reading"
|
| 594 |
)
|
|
@@ -598,6 +791,9 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="
|
|
| 598 |
with gr.Row():
|
| 599 |
current_in = gr.Slider(5, 30, value=12, step=0.5, label="Motor current (A)")
|
| 600 |
belt_speed_in = gr.Slider(0.1, 2.5, value=1.5, step=0.05, label="Belt speed (m/s)")
|
|
|
|
|
|
|
|
|
|
| 601 |
check_btn = gr.Button("Check for anomaly", variant="primary")
|
| 602 |
anomaly_output = gr.Markdown()
|
| 603 |
|
|
@@ -607,16 +803,50 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="
|
|
| 607 |
)
|
| 608 |
check_btn.click(
|
| 609 |
anomaly_fn,
|
| 610 |
-
inputs=[motor_temp_in, vibration_in, current_in, belt_speed_in],
|
| 611 |
outputs=anomaly_output,
|
| 612 |
)
|
| 613 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 614 |
with gr.Tab("📊 Model Evaluation"):
|
| 615 |
-
with gr.Accordion("
|
| 616 |
-
gr.HTML(data_badge(
|
| 617 |
-
"🧪 All charts below are computed on <b>held-out test data</b> "
|
| 618 |
-
"by <code>build_artifacts.py</code> — fully reproducible."
|
| 619 |
-
))
|
| 620 |
|
| 621 |
gr.Markdown("### 1. Intent Classifier — accuracy: **{:.1%}** · macro F1: **{:.1%}**".format(
|
| 622 |
intent_eval.get("accuracy", 0), intent_eval.get("macro_f1", 0)
|
|
@@ -643,6 +873,16 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="
|
|
| 643 |
gr.Markdown("### 4. Latency Benchmark (CPU, local components)")
|
| 644 |
gr.Image(os.path.join(ASSETS_DIR, "latency_bar.png"), show_label=False, container=False)
|
| 645 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 646 |
with gr.Accordion("📋 Full metrics tables & methodology notes", open=False):
|
| 647 |
gr.Markdown(eval_intent_section_md())
|
| 648 |
gr.Markdown(eval_anomaly_section_md())
|
|
@@ -651,7 +891,25 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="
|
|
| 651 |
gr.Markdown(EVAL_METHODOLOGY_MD)
|
| 652 |
|
| 653 |
with gr.Tab("ℹ️ About"):
|
| 654 |
-
gr.Markdown(ABOUT_MD)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 655 |
|
| 656 |
|
| 657 |
|
|
@@ -665,6 +923,6 @@ if __name__ == "__main__":
|
|
| 665 |
# the Space if so.
|
| 666 |
launch_kwargs = dict(server_name="0.0.0.0", server_port=int(os.environ.get("PORT", 7860)))
|
| 667 |
try:
|
| 668 |
-
demo.launch(**launch_kwargs,
|
| 669 |
except TypeError:
|
| 670 |
demo.launch(**launch_kwargs)
|
|
|
|
| 27 |
from src.anomaly_model import FEATURES as SENSOR_FEATURES
|
| 28 |
from src.anomaly_model import load_artifacts as load_anomaly_artifacts
|
| 29 |
from src.anomaly_model import score_reading
|
| 30 |
+
from src.alerting import send_anomaly_alert
|
| 31 |
+
from src.vision_model import generate_rack_scene, detect_boxes
|
| 32 |
from src.data_generation import generate_inventory_db, generate_orders_db
|
| 33 |
from src.intent_model import INTENT_DESCRIPTIONS, load_pipeline, predict as intent_predict
|
| 34 |
from src.inventory_db import query_inventory, query_orders
|
|
|
|
| 106 |
anomaly_eval = _load_json("anomaly_eval.json")
|
| 107 |
retrieval_eval = _load_json("retrieval_eval.json")
|
| 108 |
latency_eval = _load_json("latency_eval.json")
|
| 109 |
+
vision_eval = _load_json("vision_eval.json")
|
| 110 |
|
| 111 |
HF_TOKEN_SET = bool(_get_hf_token())
|
| 112 |
|
|
|
|
| 139 |
|
| 140 |
def chat_fn(message, history):
|
| 141 |
intent = intent_predict(intent_pipeline, message)
|
| 142 |
+
# `history` (Gradio messages format) is passed through so the LLM can handle
|
| 143 |
+
# follow-ups like "what about Zone D instead?" (bounded/cleaned in llm_client).
|
| 144 |
+
response = answer_query(message, retriever, history=history)
|
| 145 |
|
| 146 |
intent_label = INTENT_DESCRIPTIONS.get(intent.intent, intent.intent)
|
| 147 |
meta_lines = [f"**Detected intent:** {intent_label} ({intent.confidence:.0%} confidence)"]
|
|
|
|
| 220 |
# TAB 3 -- Predictive Maintenance / Anomaly Detection
|
| 221 |
# ==========================================================================
|
| 222 |
|
| 223 |
+
def anomaly_fn(motor_temp, vibration, current, belt_speed, zone_label="", send_alert=False):
|
| 224 |
reading = {
|
| 225 |
+
"motor_temp_c": motor_temp, "vibration_mm_s": vibration,
|
| 226 |
+
"current_amps": current, "belt_speed_mps": belt_speed,
|
|
|
|
|
|
|
| 227 |
}
|
| 228 |
result = score_reading(anomaly_model, anomaly_scaler, reading, threshold=anomaly_threshold)
|
| 229 |
verdict = "🔴 ANOMALY DETECTED" if result.is_anomaly else "🟢 Normal operating range"
|
| 230 |
detail = (
|
| 231 |
f"### {verdict}\n\n"
|
| 232 |
f"**Anomaly score:** {result.anomaly_score:.2f} / 1.00\n\n"
|
| 233 |
+
f"| Sensor | Value | Typical normal range |\n|---|---|---|\n"
|
|
|
|
| 234 |
f"| Motor temperature | {motor_temp:.1f} °C | 47–63 °C |\n"
|
| 235 |
f"| Vibration | {vibration:.2f} mm/s | 1.2–3.2 mm/s |\n"
|
| 236 |
f"| Motor current | {current:.1f} A | 9–15 A |\n"
|
|
|
|
| 243 |
"wear, belt misalignment, or motor overload -- schedule maintenance before "
|
| 244 |
"the next shift to avoid an unplanned stoppage."
|
| 245 |
)
|
| 246 |
+
if send_alert:
|
| 247 |
+
alert = send_anomaly_alert(zone_label or "Unspecified zone", reading, result.anomaly_score)
|
| 248 |
+
if alert.sent:
|
| 249 |
+
detail += "\n\n---\n✅ **Alert sent to Slack.**"
|
| 250 |
+
else:
|
| 251 |
+
why = alert.error or "no `SLACK_WEBHOOK_URL` secret configured"
|
| 252 |
+
detail += (f"\n\n---\n📋 **Alert preview (NOT actually sent — {why}):**\n"
|
| 253 |
+
f"```\n{alert.message_preview}\n```")
|
| 254 |
return detail
|
| 255 |
|
| 256 |
|
|
|
|
| 265 |
return ANOMALY_PRESETS[name]
|
| 266 |
|
| 267 |
|
| 268 |
+
def anomaly_batch_fn(csv_file):
|
| 269 |
+
"""CSV batch mode: run anomaly detection over many readings at once
|
| 270 |
+
(e.g. an export of the last shift's sensor logs)."""
|
| 271 |
+
if csv_file is None:
|
| 272 |
+
return "Upload a CSV first.", None, None
|
| 273 |
+
try:
|
| 274 |
+
batch_df = pd.read_csv(csv_file.name if hasattr(csv_file, "name") else csv_file)
|
| 275 |
+
except Exception as e: # noqa: BLE001
|
| 276 |
+
return f"Could not read CSV: {type(e).__name__}: {e}", None, None
|
| 277 |
+
missing = [c for c in SENSOR_FEATURES if c not in batch_df.columns]
|
| 278 |
+
if missing:
|
| 279 |
+
return (f"CSV is missing column(s): {', '.join(missing)}. Expected: "
|
| 280 |
+
f"{', '.join(SENSOR_FEATURES)} (optionally plus an `id` column).", None, None)
|
| 281 |
+
rows = []
|
| 282 |
+
for _, row in batch_df.iterrows():
|
| 283 |
+
reading = {f: float(row[f]) for f in SENSOR_FEATURES}
|
| 284 |
+
r = score_reading(anomaly_model, anomaly_scaler, reading, threshold=anomaly_threshold)
|
| 285 |
+
rows.append({**({"id": row["id"]} if "id" in batch_df.columns else {}),
|
| 286 |
+
**reading, "anomaly_score": round(r.anomaly_score, 3), "is_anomaly": r.is_anomaly})
|
| 287 |
+
out = pd.DataFrame(rows)
|
| 288 |
+
n_bad = int(out["is_anomaly"].sum())
|
| 289 |
+
summary = f"**Processed {len(out)} readings** — {n_bad} flagged anomalous ({n_bad / max(len(out), 1):.1%})."
|
| 290 |
+
path = os.path.join("/tmp", "anomaly_batch_results.csv")
|
| 291 |
+
out.to_csv(path, index=False)
|
| 292 |
+
return summary, out, path
|
| 293 |
+
|
| 294 |
+
|
| 295 |
+
SAMPLE_BATCH_CSV = os.path.join(DATA_DIR, "sample_sensor_batch.csv")
|
| 296 |
+
if not os.path.exists(SAMPLE_BATCH_CSV): # self-heal if the packaged sample is ever missing
|
| 297 |
+
import build_artifacts
|
| 298 |
+
build_artifacts.build_sample_batch_csv()
|
| 299 |
+
|
| 300 |
+
|
| 301 |
+
# ==========================================================================
|
| 302 |
+
# TAB 4 -- Visual Inventory Scan (classical computer vision)
|
| 303 |
+
# ==========================================================================
|
| 304 |
+
|
| 305 |
+
def vision_scan_fn(fill_density, noise_level, seed_state):
|
| 306 |
+
seed = int(seed_state) if seed_state is not None else 1
|
| 307 |
+
scene = generate_rack_scene(fill_prob=fill_density, seed=seed, noise_level=noise_level)
|
| 308 |
+
result = detect_boxes(scene.image)
|
| 309 |
+
diff = abs(result.detected_count - scene.true_box_count)
|
| 310 |
+
verdict = "✅ Exact match" if diff == 0 else f"⚠️ Off by {diff}"
|
| 311 |
+
summary = (
|
| 312 |
+
f"### Detected **{result.detected_count} boxes** · estimated occupancy **{result.estimated_occupancy:.0%}**\n\n"
|
| 313 |
+
f"| | CV pipeline | Ground truth |\n|---|---|---|\n"
|
| 314 |
+
f"| Box count | {result.detected_count} | {scene.true_box_count} |\n"
|
| 315 |
+
f"| Occupancy | {result.estimated_occupancy:.0%} | {scene.true_occupancy:.0%} |\n\n"
|
| 316 |
+
f"**{verdict}** — ground truth exists only because this scene is synthetic; "
|
| 317 |
+
f"a real deployment wouldn't have it."
|
| 318 |
+
)
|
| 319 |
+
return scene.image, result.annotated_image, summary, seed + 1
|
| 320 |
+
|
| 321 |
+
|
| 322 |
# ==========================================================================
|
| 323 |
# TAB 4 -- Model Evaluation
|
| 324 |
# ==========================================================================
|
|
|
|
| 391 |
systems (AS/RS), conveyor & sortation lines, AGVs/AMRs, and the WMS/WCS
|
| 392 |
software that orchestrates them — generate huge volumes of operational
|
| 393 |
data: equipment telemetry, transactions, safety logs, and ad-hoc questions
|
| 394 |
+
from floor staff. This project is a compact, end-to-end example of how an
|
| 395 |
+
AI layer can sit on top of that kind of system: routing requests correctly,
|
| 396 |
+
answering from real operational context instead of guessing, catching
|
| 397 |
+
equipment problems early, and doing all of it with **honest, reproducible
|
| 398 |
+
evaluation** rather than a demo that just "looks like it works."
|
| 399 |
|
| 400 |
## What this demonstrates
|
| 401 |
|
|
|
|
| 418 |
|
| 419 |
| Dataset | What it is | Size | How it's generated |
|
| 420 |
|---|---|---|---|
|
| 421 |
+
| **Knowledge base** | Original, hand-written articles on generic warehouse-automation concepts (AS/RS, AGV/AMR, WMS, sortation, picking strategy, predictive maintenance, safety, inventory accuracy, KPIs, energy efficiency, cross-docking, returns, RFID, dock scheduling, slotting, security, labor management) | 17 articles | Written by the developer specifically for this project |
|
| 422 |
| **Intent queries** | Example free-text operational questions/requests across 8 categories | ~480 examples | ~8 templates per category with randomised SKU codes, zone names, order IDs, and equipment IDs slotted in |
|
| 423 |
| **Inventory table** | SKU records with category, zone, on-hand units, reorder point, unit cost | 60 SKUs | Randomised within realistic ranges (fixed seed) |
|
| 424 |
| **Orders table** | Order records with status, line count, priority, zone | 80 orders | Randomised within realistic ranges (fixed seed) |
|
| 425 |
| **Sensor readings** | Conveyor/crane motor telemetry: temperature, vibration, current, belt speed | 1,000 readings (900 normal + 100 anomalous) | Normal readings drawn from realistic operating ranges; anomalies simulate known failure signatures (elevated temp/vibration/current + reduced belt speed) |
|
| 426 |
+
| **Retrieval eval set** | Hand-labelled (question → expected knowledge-base article) pairs | 17 pairs | Written by the developer to check retrieval accuracy |
|
| 427 |
|
| 428 |
## How each model works (function)
|
| 429 |
|
|
|
|
| 509 |
being honestly evaluated, rather than reaching for the biggest available
|
| 510 |
model by default.
|
| 511 |
|
| 512 |
+
## Extended capabilities
|
| 513 |
+
|
| 514 |
+
- **Multi-turn conversation** — the assistant keeps the last few turns, so follow-ups like *"what about Zone D instead?"* work.
|
| 515 |
+
- **📷 Visual Inventory Scan** — a classical (non-deep-learning) computer-vision pipeline counts boxes and estimates occupancy on synthetic rack images. Evaluated honestly: **100%** on clean renders, but it degrades as simulated photographic noise rises, so the headline number is reported at a moderate noise level (see Model Evaluation).
|
| 516 |
+
- **Batch CSV scoring** — upload a CSV of sensor readings (a sample is provided) to score a whole shift's logs at once, and download the results.
|
| 517 |
+
- **Slack alerting** — anomalies can trigger a Slack alert. Real if a `SLACK_WEBHOOK_URL` secret is set; otherwise a clearly-labelled preview that is *never* presented as sent.
|
| 518 |
+
- **Bilingual UI** — switch the explanatory text between English and 日本語 with the selector at the top. (Charts and metric tables stay in English; the assistant answers in whatever language you ask in.)
|
| 519 |
+
|
| 520 |
## Limitations & next steps
|
| 521 |
|
| 522 |
- All data here is **synthetic**, for portfolio/demo purposes — a production
|
| 523 |
version would connect to real WMS/WCS APIs and historical sensor logs.
|
| 524 |
+
- The intent set (8 classes) and knowledge base (17 articles) are intentionally
|
| 525 |
small to keep the demo fast and auditable; both are easy to extend.
|
| 526 |
- The anomaly detector uses 4 hand-picked features; a production system would
|
| 527 |
likely use a richer multivariate sensor set and a supervised or
|
|
|
|
| 552 |
.data-badge b { color: #1e3a8a; }
|
| 553 |
"""
|
| 554 |
|
| 555 |
+
|
| 556 |
+
# ==========================================================================
|
| 557 |
+
# Bilingual (English / 日本語) static content
|
| 558 |
+
# ==========================================================================
|
| 559 |
+
# Static explanatory text (subtitle, per-tab data/function panels, About tab)
|
| 560 |
+
# is switchable via the language selector. Dynamic content (charts, metric
|
| 561 |
+
# tables, the LLM's answers) stays in English; the assistant itself will
|
| 562 |
+
# answer in whatever language the question is asked in, since the hosted LLM
|
| 563 |
+
# is multilingual.
|
| 564 |
+
|
| 565 |
+
from src.knowledge_base import KNOWLEDGE_BASE
|
| 566 |
+
_N_KB = len(KNOWLEDGE_BASE)
|
| 567 |
+
I18N = {
|
| 568 |
+
"subtitle": {
|
| 569 |
+
"en": "<p style='text-align:center; color:gray;'>LLM-powered intralogistics copilot · intent routing · "
|
| 570 |
+
"multi-turn RAG · predictive maintenance · computer vision</p>",
|
| 571 |
+
"ja": "<p style='text-align:center; color:gray;'>LLM搭載の物流オペレーション支援AI · 意図分類 · "
|
| 572 |
+
"マルチターン対話(RAG) · 予知保全 · コンピュータビジョン</p>",
|
| 573 |
+
},
|
| 574 |
+
"badge_assistant": {
|
| 575 |
+
"en": "🧪 <b>Source:</b> hand-written by the developer — %d original knowledge-base articles on warehouse "
|
| 576 |
+
"operations (AS/RS, AGV/AMR, WMS, sortation, picking, maintenance, safety, cross-docking, RFID, slotting, "
|
| 577 |
+
"labor management…) plus ~480 template-generated example queries. Not scraped or sourced from any real company.<br>"
|
| 578 |
+
"<b>What it describes:</b> generic intralogistics/automation concepts and operational scenarios (equipment "
|
| 579 |
+
"faults, safety incidents, order/inventory questions) — general domain knowledge, not any specific facility's live data.<br>"
|
| 580 |
+
"<b>Function:</b> answers free-text warehouse-ops questions by classifying the query's intent, retrieving the "
|
| 581 |
+
"most relevant knowledge-base passage(s) (RAG), then generating a grounded answer with a hosted LLM. It keeps the "
|
| 582 |
+
"last few turns of the conversation, so follow-ups like “what about Zone D instead?” work. If no LLM is available, "
|
| 583 |
+
"it falls back to showing the retrieved passages directly." % _N_KB,
|
| 584 |
+
"ja": "🧪 <b>データの出典:</b>開発者が手書きした倉庫運用ナレッジ記事%d本(AS/RS、AGV/AMR、WMS、仕分け、ピッキング、保全、安全、"
|
| 585 |
+
"クロスドッキング、RFID、スロッティング、労務管理など)と、テンプレートから生成した約480件の問い合わせ例。"
|
| 586 |
+
"実在企業のデータは使用していません。<br>"
|
| 587 |
+
"<b>内容:</b>物流・自動化に関する一般的な概念と運用シナリオ(設備故障、安全インシデント、在庫・注文の問い合わせ等)であり、"
|
| 588 |
+
"特定施設のリアルタイムデータではありません。<br>"
|
| 589 |
+
"<b>機能:</b>自由入力の質問に対して、①意図分類 → ②ナレッジ検索(RAG)→ ③ホスト型LLMによる回答生成の順に処理します。"
|
| 590 |
+
"直近の会話履歴も保持するため、「Zone Dの場合は?」のような追加質問にも対応します。LLMが利用できない場合は、"
|
| 591 |
+
"検索結果をそのまま表示します。" % _N_KB,
|
| 592 |
+
},
|
| 593 |
+
"badge_inventory": {
|
| 594 |
+
"en": f"🧪 <b>Source:</b> randomly generated by the developer's code (<code>src/data_generation.py</code>, fixed seed) — "
|
| 595 |
+
f"{len(inventory_df)} synthetic SKU records and {len(orders_df)} synthetic orders. Not exported from any real company's WMS.<br>"
|
| 596 |
+
"<b>What it describes:</b> a stand-in inventory table (SKU, category, zone, on-hand units, reorder point, unit cost) and "
|
| 597 |
+
"orders table (order ID, status, line count, priority, zone) — realistic in shape, but fictional records.<br>"
|
| 598 |
+
"<b>Function:</b> classifies whether a query is about inventory or order status, extracts SKU / order-ID / zone "
|
| 599 |
+
"identifiers with regex, and filters the matching table — a lightweight stand-in for a natural-language WMS query tool.",
|
| 600 |
+
"ja": f"🧪 <b>データの出典:</b>開発者のコード(<code>src/data_generation.py</code>、乱数シード固定)で生成した合成データ"
|
| 601 |
+
f"(SKU {len(inventory_df)}件、注文 {len(orders_df)}件)。実在企業のWMSから出力したものではありません。<br>"
|
| 602 |
+
"<b>内容:</b>在庫テーブル(SKU・カテゴリ・ゾーン・在庫数・発注点・単価)と注文テーブル(注文ID・ステータス・明細数・優先度・ゾーン)。"
|
| 603 |
+
"形式は現実的ですが、内容は架空のレコードです。<br>"
|
| 604 |
+
"<b>機能:</b>問い合わせが在庫か注文状況かを分類し、SKU・注文ID・ゾーンを正規表現で抽出して該当テーブルを絞り込みます。"
|
| 605 |
+
"自然言語によるWMS照会ツールの簡易版です。",
|
| 606 |
+
},
|
| 607 |
+
"badge_predictive": {
|
| 608 |
+
"en": "🧪 <b>Source:</b> randomly generated by the developer's code (<code>src/data_generation.py</code>, fixed seed) — 1,000 synthetic "
|
| 609 |
+
"sensor readings (900 normal + 100 simulated-fault). Not logged from real equipment.<br>"
|
| 610 |
+
"<b>What it describes:</b> conveyor/crane motor readings — temperature, vibration, current draw, belt speed — with 'anomaly' "
|
| 611 |
+
"readings modelled on failure signatures (elevated temp/vibration/current with reduced belt speed, as with bearing wear or overload).<br>"
|
| 612 |
+
"<b>Function:</b> an Isolation Forest is fit unsupervised (it never sees an anomaly label while fitting); its decision threshold "
|
| 613 |
+
"is then calibrated on a small labelled calibration set (never on test data). Check one reading with the sliders, or upload a CSV "
|
| 614 |
+
"to score a whole shift's logs. Anomalies can trigger a Slack alert (real if a <code>SLACK_WEBHOOK_URL</code> secret is set, "
|
| 615 |
+
"otherwise a clearly-labelled preview).",
|
| 616 |
+
"ja": "🧪 <b>データの出典:</b>開発者のコード(<code>src/data_generation.py</code>、乱数シード固定)で生成した合成センサーデータ1,000件"
|
| 617 |
+
"(正常900+故障模擬100)。実機のログではありません。<br>"
|
| 618 |
+
"<b>内容:</b>コンベヤ/クレーンのモーター計測値(温度・振動・電流・ベルト速度)。異常データは、ベアリング摩耗や過負荷に見られる"
|
| 619 |
+
"故障パターン(温度・振動・電流の上昇とベルト速度の低下)を模して作成しています。<br>"
|
| 620 |
+
"<b>機能:</b>Isolation Forestを教師なしで学習し(学習時に異常ラベルは使用しません)、判定しきい値は少量のラベル付き較正データで調整します"
|
| 621 |
+
"(テストデータは使用しません)。スライダーで1件を判定するほか、CSVをアップロードしてシフト全体のログを一括判定できます。"
|
| 622 |
+
"異常時はSlack通知が可能です(<code>SLACK_WEBHOOK_URL</code>設定時は実送信、未設定時はプレビュー表示)。",
|
| 623 |
+
},
|
| 624 |
+
"badge_vision": {
|
| 625 |
+
"en": "🧪 <b>Source:</b> synthetic rack-scene images generated by the developer's code (<code>src/vision_model.py</code>) — not real "
|
| 626 |
+
"warehouse photos (none were available to license).<br>"
|
| 627 |
+
"<b>What it describes:</b> a simplified front view of a 3-row × 6-column storage rack with randomly placed box-like shapes "
|
| 628 |
+
"standing in for pallets/cartons, plus adjustable simulated photographic noise.<br>"
|
| 629 |
+
"<b>Function:</b> a <i>classical</i> (non-deep-learning) computer-vision pipeline — colour-based background/shelf segmentation, "
|
| 630 |
+
"morphological cleanup, connected components, geometric filtering — counts boxes and estimates rack occupancy. Deliberately the "
|
| 631 |
+
"opposite of a pretrained detector: zero model downloads, fully explainable, instant on CPU. The Model Evaluation tab shows how "
|
| 632 |
+
"accuracy degrades as noise rises.",
|
| 633 |
+
"ja": "🧪 <b>データの出典:</b>開発者のコード(<code>src/vision_model.py</code>)で生成した合成のラック画像。実際の倉庫写真ではありません"
|
| 634 |
+
"(利用可能なライセンス素材がないため)。<br>"
|
| 635 |
+
"<b>内容:</b>3段×6列の保管ラックを正面から見た簡易画像で、パレット/ケースに見立てた箱がランダムに配置されます。"
|
| 636 |
+
"撮影ノイズの強さも調整できます。<br>"
|
| 637 |
+
"<b>機能:</b>深層学習を使わない<i>古典的</i>な画像処理パイプライン(色によるラック背景の分離 → モルフォロジー処理 → 連結成分抽出 → "
|
| 638 |
+
"形状フィルタ)で、箱の数と棚の占有率を推定します。学習済みモデルのダウンロードが不要で、処理過程を説明でき、CPUで即時に動作します。"
|
| 639 |
+
"ノイズ増加時の精度低下は「Model Evaluation」タブで確認できます。",
|
| 640 |
+
},
|
| 641 |
+
"badge_eval": {
|
| 642 |
+
"en": "🧪 All charts below are computed on <b>held-out test data</b> by <code>build_artifacts.py</code> — fully reproducible, not cherry-picked.",
|
| 643 |
+
"ja": "🧪 以下の図表はすべて、<code>build_artifacts.py</code>が<b>ホールドアウトしたテストデータ</b>で算出したものです。再現可能で、結果の取捨選択は行っていません。",
|
| 644 |
+
},
|
| 645 |
+
}
|
| 646 |
+
|
| 647 |
+
|
| 648 |
+
def about_ja() -> str:
|
| 649 |
+
return f"""
|
| 650 |
+
# 🏭 スマート倉庫AIアシスタント
|
| 651 |
+
|
| 652 |
+
**物流・倉庫自動化の運用を題材に、実務的なAI活用を示すポートフォリオプロジェクトです。**
|
| 653 |
+
|
| 654 |
+
AS/RS、コンベヤ・仕分け、AGV/AMR、WMS/WCSといった倉庫自動化の現場では、設備テレメトリ、取引データ、安全記録、現場からの問い合わせなど
|
| 655 |
+
大量の運用データが生まれます。本プロジェクトは、その上にAIレイヤーを載せる最小構成の実装例です。依頼を正しく振り分け、推測ではなく
|
| 656 |
+
根拠のある情報で回答し、設備の異常を早期に検出し、さらに**再現可能で誠実な評価**を行うことを重視しています。
|
| 657 |
+
|
| 658 |
+
## 機能一覧
|
| 659 |
+
|
| 660 |
+
| 機能 | タブ | 主な技術 |
|
| 661 |
+
|---|---|---|
|
| 662 |
+
| ナレッジに基づく**LLM対話アシスタント**(RAG/マルチターン会話対応) | AI Assistant | TF-IDF検索 + ホスト型LLM |
|
| 663 |
+
| **意図分類**による問い合わせの振り分け(8カテゴリ) | AI Assistant / Inventory | TF-IDF + ロジスティック回帰 |
|
| 664 |
+
| 在庫・注文の**自然文照会** | Inventory & Order Query | 正規表現によるスロット抽出 + 意図分類 |
|
| 665 |
+
| センサーデータの**予知保全**(1件判定/CSV一括判定/Slack通知) | Predictive Maintenance | Isolation Forest(教師なし)+ 較正済みしきい値 |
|
| 666 |
+
| ラック画像の**箱数カウントと占有率推定** | Visual Inventory Scan | 古典的画像処理(OpenCV) |
|
| 667 |
+
| 全コンポーネントの**再現可能な評価** | Model Evaluation | scikit-learn指標、ホールドアウト検証 |
|
| 668 |
+
|
| 669 |
+
## データセットの概要
|
| 670 |
+
|
| 671 |
+
**本プロジェクトのデータはすべて、自作コードによる合成データです。**実在企業のシステムからの取得・流用は一切ありません。
|
| 672 |
+
プライバシーやライセンスの懸念なく、誰でも再現・共有できます。
|
| 673 |
+
|
| 674 |
+
| データ | 内容 | 件数 | 生成方法 |
|
| 675 |
+
|---|---|---|---|
|
| 676 |
+
| ナレッジベース | 倉庫自動化の一般的概念を解説した手書き記事 | {_N_KB}記事 | 本プロジェクトのために作成 |
|
| 677 |
+
| 意図分類クエリ | 8カテゴリの問い合わせ例 | 約480件 | テンプレートにSKU・ゾーン・設備IDなどをランダムに埋め込み |
|
| 678 |
+
| 在庫テーブル | SKU・カテゴリ・ゾーン・在庫数・発注点・単価 | 60 SKU | 現実的な範囲でランダム生成(シード固定) |
|
| 679 |
+
| 注文テーブル | ステータス・明細数・優先度・ゾーン | 80件 | 現実的な範囲でランダム生成(シード固定) |
|
| 680 |
+
| センサーデータ | 温度・振動・電流・ベルト速度 | 1,000件(正常900/異常100) | 正常は運転範囲から、異常は故障パターンを模して生成 |
|
| 681 |
+
| ラック画像 | 箱の配置が既知の合成画像 | 随時生成 | PILで描画。ノイズ強度を調整可能 |
|
| 682 |
+
| 検索評価セット | 質問と正解記事のペア | {_N_KB}組 | 検索精度確認のため手作業で作成 |
|
| 683 |
+
|
| 684 |
+
## 各モデルの役割
|
| 685 |
+
|
| 686 |
+
| モデル | 目的 | アルゴリズム | 入力 → 出力 |
|
| 687 |
+
|---|---|---|---|
|
| 688 |
+
| 意図分類器 | 問い合わせの種類を判定 | TF-IDF + ロジスティック回帰 | 文章 → 意図ラベル + 確信度 |
|
| 689 |
+
| 検索(RAG) | 回答の根拠となる記事を取得 | TF-IDF + コサイン類似度 | 文章 → 関連記事の上位k件 |
|
| 690 |
+
| LLMアシスタント | 根拠に基づく自然な回答を生成 | ホスト型LLM(複数モデルのフォールバック) | 質問 + 根拠 + 会話履歴 → 回答 |
|
| 691 |
+
| 在庫・注文照会 | 質問をテーブル検索に変換 | 正規表現 + 意図分類 | 文章 → 絞り込み結果 |
|
| 692 |
+
| 異常検知 | 設備の異常な計測値を早期に検出 | Isolation Forest + 較正済みしきい値 | 4つのセンサー値 → 正常/異常 + スコア |
|
| 693 |
+
| ラックスキャナ | 箱の数と占有率を推定 | 色分離 + 輪郭抽出(古典的画像処理) | 画像 → 箱数 + 占有率 + 検出枠 |
|
| 694 |
+
|
| 695 |
+
## 試し方
|
| 696 |
+
|
| 697 |
+
1. **AI Assistant** — 例の質問を選ぶか自由に入力します。続けて「Zone Dの場合は?」などと尋ねると、直前の文脈を踏まえて回答します。
|
| 698 |
+
2. **Inventory & Order Query** — 「SKU-1042 は Zone B に何個ある?」「注文 #10007 の状況は?」などを試します。
|
| 699 |
+
3. **Predictive Maintenance** — プリセットまたはスライダーで1件を判定します。「CSV一括判定」でサンプルCSVをダウンロードし、そのままアップロードして一括判定も試せます。
|
| 700 |
+
4. **Visual Inventory Scan** — 「新しいラック画像を生成」を押し、ノイズを上げて検出が難しくなる様子を確認します。
|
| 701 |
+
5. **Model Evaluation** — すべての図表はテストデータで算出しており、`build_artifacts.py`を再実行すれば同じ数値を再現できます。
|
| 702 |
+
|
| 703 |
+
## 評価の信頼性について
|
| 704 |
+
|
| 705 |
+
開発中に2件のデータリーク(意図分類での学習/テスト間の文面重複、異常検知でのしきい値への正解率の混入)を発見・修正しました。
|
| 706 |
+
修正後の指標は、意図分類 正解率 {intent_eval.get('accuracy', 0):.1%}、異常検知 F1 {anomaly_eval.get('f1', 0):.1%} です。
|
| 707 |
+
画像パイプラインも、ノイズなしでは100%となるため、ノイズを加えた条件(標準ノイズ0.4で完全一致率 {vision_eval.get('exact_match_rate', 0):.0%})を報告しています。
|
| 708 |
+
|
| 709 |
+
## 制約と今後の課題
|
| 710 |
+
|
| 711 |
+
- データはすべて**合成**です。実運用ではWMS/WCSのAPIや実際のセンサーログに接続します。
|
| 712 |
+
- 画像は単純化した合成ラックであり、実写真への適用には学習済み検出器(YOLO等)の微調整が現実的です。
|
| 713 |
+
- Slack通知は`SLACK_WEBHOOK_URL`未設定の場合はプレビューのみで、実際には送信されません。
|
| 714 |
+
"""
|
| 715 |
+
|
| 716 |
+
|
| 717 |
def data_badge(html: str) -> str:
|
| 718 |
return f'<div class="data-badge">{html}</div>'
|
| 719 |
|
| 720 |
with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=CUSTOM_CSS, title="Smart Warehouse AI Assistant") as demo:
|
| 721 |
+
gr.Markdown("<h1 id='title-banner'>🏭 Smart Warehouse AI Assistant</h1>")
|
| 722 |
+
subtitle_md = gr.Markdown(I18N["subtitle"]["en"])
|
| 723 |
+
lang_radio = gr.Radio(["English", "日本語"], value="English", label="🌐 Language / 言語", interactive=True)
|
|
|
|
|
|
|
| 724 |
if not HF_TOKEN_SET:
|
| 725 |
gr.Markdown(
|
| 726 |
"> ⚠️ **No HF token secret detected.** The AI Assistant tab will run in "
|
|
|
|
| 730 |
)
|
| 731 |
|
| 732 |
with gr.Tab("💬 AI Assistant"):
|
| 733 |
+
with gr.Accordion("🧪 About this data & function / データ・機能について", open=False):
|
| 734 |
+
badge_assistant = gr.HTML(data_badge(I18N["badge_assistant"]["en"]))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 735 |
gr.ChatInterface(
|
| 736 |
fn=chat_fn,
|
| 737 |
type="messages",
|
|
|
|
| 750 |
test_llm_btn.click(test_llm_fn, inputs=None, outputs=test_llm_output)
|
| 751 |
|
| 752 |
with gr.Tab("📦 Inventory & Order Query"):
|
| 753 |
+
with gr.Accordion("🧪 About this data & function / データ・機能について", open=False):
|
| 754 |
+
badge_inventory = gr.HTML(data_badge(I18N["badge_inventory"]["en"]))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 755 |
with gr.Row():
|
| 756 |
+
inv_input = gr.Textbox(label="Query", placeholder="How many units of SKU-1042 are in Zone B?", scale=4)
|
| 757 |
inv_btn = gr.Button("Search", variant="primary", scale=1)
|
| 758 |
inv_note = gr.Markdown()
|
| 759 |
inv_result = gr.Dataframe(label="Results", wrap=True)
|
|
|
|
| 762 |
|
| 763 |
gr.Examples(
|
| 764 |
examples=[
|
| 765 |
+
"How many units of SKU-1042 are in Zone B?",
|
| 766 |
"What's the status of order #10007?",
|
| 767 |
"Show me low stock items",
|
| 768 |
"Any delayed orders?",
|
| 769 |
+
"Is SKU-1015 in stock at Zone A?",
|
| 770 |
"Do we have enough SKU-1030 to fulfill 200 units?",
|
| 771 |
"Track order #10021 for me",
|
| 772 |
"Show the fulfillment status of #10045",
|
|
|
|
| 780 |
gr.Dataframe(orders_df, wrap=True)
|
| 781 |
|
| 782 |
with gr.Tab("⚠️ Predictive Maintenance"):
|
| 783 |
+
with gr.Accordion("🧪 About this data & function / データ・機能について", open=False):
|
| 784 |
+
badge_predictive = gr.HTML(data_badge(I18N["badge_predictive"]["en"]))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 785 |
preset_dropdown = gr.Dropdown(
|
| 786 |
choices=list(ANOMALY_PRESETS.keys()), label="Load a preset reading", value="Normal reading"
|
| 787 |
)
|
|
|
|
| 791 |
with gr.Row():
|
| 792 |
current_in = gr.Slider(5, 30, value=12, step=0.5, label="Motor current (A)")
|
| 793 |
belt_speed_in = gr.Slider(0.1, 2.5, value=1.5, step=0.05, label="Belt speed (m/s)")
|
| 794 |
+
with gr.Row():
|
| 795 |
+
zone_label_in = gr.Textbox(label="Equipment / zone label (used in alerts)", placeholder="e.g. Conveyor-14, Zone C", scale=3)
|
| 796 |
+
send_alert_in = gr.Checkbox(label="Send Slack alert if anomalous", value=False, scale=1)
|
| 797 |
check_btn = gr.Button("Check for anomaly", variant="primary")
|
| 798 |
anomaly_output = gr.Markdown()
|
| 799 |
|
|
|
|
| 803 |
)
|
| 804 |
check_btn.click(
|
| 805 |
anomaly_fn,
|
| 806 |
+
inputs=[motor_temp_in, vibration_in, current_in, belt_speed_in, zone_label_in, send_alert_in],
|
| 807 |
outputs=anomaly_output,
|
| 808 |
)
|
| 809 |
|
| 810 |
+
with gr.Accordion("📋 Batch mode: upload a CSV of sensor readings", open=False):
|
| 811 |
+
gr.Markdown(
|
| 812 |
+
f"Upload a CSV with columns `{'`, `'.join(SENSOR_FEATURES)}` (optionally an `id` column) to run "
|
| 813 |
+
"detection over many readings at once — e.g. an export of the last shift's sensor logs. "
|
| 814 |
+
"Don't have one? Download the sample below and upload it back."
|
| 815 |
+
)
|
| 816 |
+
gr.File(value=SAMPLE_BATCH_CSV, label="Sample CSV (40 readings, ~15% faulty)", interactive=False)
|
| 817 |
+
batch_csv_in = gr.File(label="Your sensor readings CSV", file_types=[".csv"])
|
| 818 |
+
batch_btn = gr.Button("Run batch detection")
|
| 819 |
+
batch_summary = gr.Markdown()
|
| 820 |
+
batch_results = gr.Dataframe(label="Results", wrap=True)
|
| 821 |
+
batch_download = gr.File(label="Download results CSV")
|
| 822 |
+
batch_btn.click(anomaly_batch_fn, inputs=batch_csv_in,
|
| 823 |
+
outputs=[batch_summary, batch_results, batch_download])
|
| 824 |
+
|
| 825 |
+
with gr.Tab("📷 Visual Inventory Scan"):
|
| 826 |
+
with gr.Accordion("🧪 About this data & function / データ・機能について", open=False):
|
| 827 |
+
badge_vision = gr.HTML(data_badge(I18N["badge_vision"]["en"]))
|
| 828 |
+
gr.Markdown(
|
| 829 |
+
"Generate a synthetic rack photo and watch the pipeline count boxes and estimate "
|
| 830 |
+
"occupancy. Raise the noise slider to see detection get harder, the way a dim or "
|
| 831 |
+
"grainy real photo would."
|
| 832 |
+
)
|
| 833 |
+
with gr.Row():
|
| 834 |
+
fill_density_in = gr.Slider(0.1, 0.95, value=0.6, step=0.05, label="Rack fill density (scene generation)")
|
| 835 |
+
noise_in = gr.Slider(0.0, 1.0, value=0.3, step=0.05, label="Simulated photographic noise")
|
| 836 |
+
vision_seed_state = gr.State(value=1)
|
| 837 |
+
new_scene_btn = gr.Button("🎲 Generate new rack scene", variant="primary")
|
| 838 |
+
with gr.Row():
|
| 839 |
+
vision_raw_img = gr.Image(label="Rack photo (input)", type="pil")
|
| 840 |
+
vision_annotated_img = gr.Image(label="Detected boxes (output)", type="pil")
|
| 841 |
+
vision_summary = gr.Markdown()
|
| 842 |
+
_vision_io = dict(inputs=[fill_density_in, noise_in, vision_seed_state],
|
| 843 |
+
outputs=[vision_raw_img, vision_annotated_img, vision_summary, vision_seed_state])
|
| 844 |
+
new_scene_btn.click(vision_scan_fn, **_vision_io)
|
| 845 |
+
demo.load(vision_scan_fn, **_vision_io)
|
| 846 |
+
|
| 847 |
with gr.Tab("📊 Model Evaluation"):
|
| 848 |
+
with gr.Accordion("🧪 About this data & function / データ・機能について", open=False):
|
| 849 |
+
badge_eval = gr.HTML(data_badge(I18N["badge_eval"]["en"]))
|
|
|
|
|
|
|
|
|
|
| 850 |
|
| 851 |
gr.Markdown("### 1. Intent Classifier — accuracy: **{:.1%}** · macro F1: **{:.1%}**".format(
|
| 852 |
intent_eval.get("accuracy", 0), intent_eval.get("macro_f1", 0)
|
|
|
|
| 873 |
gr.Markdown("### 4. Latency Benchmark (CPU, local components)")
|
| 874 |
gr.Image(os.path.join(ASSETS_DIR, "latency_bar.png"), show_label=False, container=False)
|
| 875 |
|
| 876 |
+
gr.Markdown("### 5. Visual Inventory Scan — exact box-count match: **{:.0%}** (noise={}) · count MAE: **{:.2f}** boxes".format(
|
| 877 |
+
vision_eval.get("exact_match_rate", 0), vision_eval.get("standard_noise_level", "?"), vision_eval.get("count_mae", 0)
|
| 878 |
+
))
|
| 879 |
+
with gr.Row():
|
| 880 |
+
gr.Image(os.path.join(ASSETS_DIR, "vision_metrics_bar.png"), show_label=False, container=False)
|
| 881 |
+
gr.Image(os.path.join(ASSETS_DIR, "vision_noise_robustness.png"), show_label=False, container=False)
|
| 882 |
+
with gr.Row():
|
| 883 |
+
gr.Image(os.path.join(ASSETS_DIR, "vision_sample_raw.png"), label="Sample input", container=True)
|
| 884 |
+
gr.Image(os.path.join(ASSETS_DIR, "vision_sample_annotated.png"), label="Sample detection", container=True)
|
| 885 |
+
|
| 886 |
with gr.Accordion("📋 Full metrics tables & methodology notes", open=False):
|
| 887 |
gr.Markdown(eval_intent_section_md())
|
| 888 |
gr.Markdown(eval_anomaly_section_md())
|
|
|
|
| 891 |
gr.Markdown(EVAL_METHODOLOGY_MD)
|
| 892 |
|
| 893 |
with gr.Tab("ℹ️ About"):
|
| 894 |
+
about_md = gr.Markdown(ABOUT_MD)
|
| 895 |
+
|
| 896 |
+
def set_language(lang):
|
| 897 |
+
k = "ja" if lang == "日本語" else "en"
|
| 898 |
+
return (
|
| 899 |
+
I18N["subtitle"][k],
|
| 900 |
+
data_badge(I18N["badge_assistant"][k]),
|
| 901 |
+
data_badge(I18N["badge_inventory"][k]),
|
| 902 |
+
data_badge(I18N["badge_predictive"][k]),
|
| 903 |
+
data_badge(I18N["badge_vision"][k]),
|
| 904 |
+
data_badge(I18N["badge_eval"][k]),
|
| 905 |
+
about_ja() if k == "ja" else ABOUT_MD,
|
| 906 |
+
)
|
| 907 |
+
|
| 908 |
+
lang_radio.change(
|
| 909 |
+
set_language, inputs=lang_radio,
|
| 910 |
+
outputs=[subtitle_md, badge_assistant, badge_inventory, badge_predictive,
|
| 911 |
+
badge_vision, badge_eval, about_md],
|
| 912 |
+
)
|
| 913 |
|
| 914 |
|
| 915 |
|
|
|
|
| 923 |
# the Space if so.
|
| 924 |
launch_kwargs = dict(server_name="0.0.0.0", server_port=int(os.environ.get("PORT", 7860)))
|
| 925 |
try:
|
| 926 |
+
demo.launch(**launch_kwargs, ssr_mode=False)
|
| 927 |
except TypeError:
|
| 928 |
demo.launch(**launch_kwargs)
|
assets/anomaly_confusion_matrix.png
CHANGED
|
|
assets/anomaly_metrics_bar.png
CHANGED
|
|
assets/anomaly_roc_curve.png
CHANGED
|
|
assets/intent_confusion_matrix.png
CHANGED
|
|
Git LFS Details
|
assets/intent_dataset_composition.png
CHANGED
|
|
assets/intent_per_class_bar.png
CHANGED
|
|
assets/latency_bar.png
CHANGED
|
|
assets/retrieval_hitrate_bar.png
CHANGED
|
|
assets/sensor_distributions.png
CHANGED
|
|
Git LFS Details
|
assets/vision_metrics_bar.png
ADDED
|
assets/vision_noise_robustness.png
ADDED
|
assets/vision_sample_annotated.png
ADDED
|
Git LFS Details
|
assets/vision_sample_raw.png
ADDED
|
Git LFS Details
|
build_artifacts.py
CHANGED
|
@@ -446,10 +446,83 @@ def build_latency_benchmark(intent_metrics, anomaly_metrics):
|
|
| 446 |
plt.close(fig)
|
| 447 |
|
| 448 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 449 |
if __name__ == "__main__":
|
| 450 |
intent_metrics = build_intent_classifier()
|
| 451 |
anomaly_metrics = build_anomaly_detector()
|
| 452 |
build_retrieval_eval()
|
| 453 |
build_inventory_and_orders()
|
| 454 |
build_latency_benchmark(intent_metrics, anomaly_metrics)
|
|
|
|
|
|
|
| 455 |
print("\nAll artifacts built successfully.")
|
|
|
|
| 446 |
plt.close(fig)
|
| 447 |
|
| 448 |
|
| 449 |
+
def build_sample_batch_csv():
|
| 450 |
+
"""A small downloadable CSV (40 readings, ~15% faulty) for demoing the app's
|
| 451 |
+
batch-upload mode. Fresh seed -> not drawn from the model's training data."""
|
| 452 |
+
from src.data_generation import generate_sensor_dataset
|
| 453 |
+
df = generate_sensor_dataset(n_normal=34, n_anomaly=6, seed=7)
|
| 454 |
+
out = df[FEATURES].round(2)
|
| 455 |
+
out.insert(0, "id", [f"R{i + 1:03d}" for i in range(len(out))])
|
| 456 |
+
out.to_csv(os.path.join(DATA_DIR, "sample_sensor_batch.csv"), index=False)
|
| 457 |
+
print(f"sample batch CSV written ({len(out)} rows)")
|
| 458 |
+
|
| 459 |
+
|
| 460 |
+
def build_vision_evaluation():
|
| 461 |
+
"""Evaluate the classical-CV rack scanner on synthetic scenes with known ground
|
| 462 |
+
truth: headline metrics at a moderate noise level (0.4) plus a noise sweep, so
|
| 463 |
+
the result isn't a suspiciously perfect number measured on clean renders only."""
|
| 464 |
+
print("== Vision: rack box-counting / occupancy ==")
|
| 465 |
+
from src.vision_model import generate_rack_scene, detect_boxes
|
| 466 |
+
|
| 467 |
+
STD, N = 0.4, 100
|
| 468 |
+
exact, cerr, oerr = 0, [], []
|
| 469 |
+
for i in range(N):
|
| 470 |
+
sc = generate_rack_scene(0.25 + (i % 8) * 0.08, 1000 + i, STD)
|
| 471 |
+
r = detect_boxes(sc.image)
|
| 472 |
+
exact += int(r.detected_count == sc.true_box_count)
|
| 473 |
+
cerr.append(abs(r.detected_count - sc.true_box_count))
|
| 474 |
+
oerr.append(abs(r.estimated_occupancy - sc.true_occupancy))
|
| 475 |
+
exact_rate, count_mae, occ_mae = exact / N, float(np.mean(cerr)), float(np.mean(oerr))
|
| 476 |
+
print(f"[noise={STD}] exact_match={exact_rate:.3f} count_MAE={count_mae:.3f} occ_MAE={occ_mae:.4f}")
|
| 477 |
+
|
| 478 |
+
levels = [0.0, 0.2, 0.4, 0.6, 0.8, 1.0]
|
| 479 |
+
sweep = []
|
| 480 |
+
for nz in levels:
|
| 481 |
+
n, m = 60, 0
|
| 482 |
+
for i in range(n):
|
| 483 |
+
sc = generate_rack_scene(0.3 + (i % 7) * 0.1, 2000 + i, nz)
|
| 484 |
+
m += int(detect_boxes(sc.image).detected_count == sc.true_box_count)
|
| 485 |
+
sweep.append(m / n)
|
| 486 |
+
print(f" sweep noise={nz:.1f} exact_match={m / n:.3f}")
|
| 487 |
+
|
| 488 |
+
fig, ax = plt.subplots(figsize=(6.5, 4.5))
|
| 489 |
+
ax.plot(levels, [v * 100 for v in sweep], marker="o", color="#7c3aed", linewidth=2, markersize=7)
|
| 490 |
+
ax.set_xlabel("Simulated photographic noise level"); ax.set_ylabel("Exact box-count match rate (%)")
|
| 491 |
+
ax.set_title("Vision Pipeline: Accuracy vs. Noise Level"); ax.set_ylim(0, 105); ax.grid(alpha=0.3)
|
| 492 |
+
fig.tight_layout(); fig.savefig(os.path.join(ASSETS_DIR, "vision_noise_robustness.png"), dpi=150); plt.close(fig)
|
| 493 |
+
|
| 494 |
+
# three separately-scaled panels (percent / boxes / percentage points are not comparable)
|
| 495 |
+
fig, axes = plt.subplots(1, 3, figsize=(9, 4))
|
| 496 |
+
panels = [("Exact count\nmatch rate", exact_rate * 100, "%", 100, "#7c3aed"),
|
| 497 |
+
("Count MAE", count_mae, "boxes", max(1, count_mae * 2), "#f59e0b"),
|
| 498 |
+
("Occupancy MAE", occ_mae * 100, "pp", max(5, occ_mae * 200), "#10b981")]
|
| 499 |
+
for ax, (name, val, unit, ymax, color) in zip(axes, panels):
|
| 500 |
+
ax.bar([name], [val], color=color, width=0.5); ax.set_ylim(0, ymax * 1.25); ax.set_ylabel(unit)
|
| 501 |
+
ax.text(0, val, f"{val:.2f}", ha="center", va="bottom", fontsize=11, fontweight="bold")
|
| 502 |
+
ax.grid(axis="y", alpha=0.3)
|
| 503 |
+
fig.suptitle(f"Vision Pipeline Metrics (noise={STD})")
|
| 504 |
+
fig.tight_layout(); fig.savefig(os.path.join(ASSETS_DIR, "vision_metrics_bar.png"), dpi=150); plt.close(fig)
|
| 505 |
+
|
| 506 |
+
demo_scene = generate_rack_scene(0.55, 42, STD)
|
| 507 |
+
detect_boxes(demo_scene.image).annotated_image.save(os.path.join(ASSETS_DIR, "vision_sample_annotated.png"))
|
| 508 |
+
demo_scene.image.save(os.path.join(ASSETS_DIR, "vision_sample_raw.png"))
|
| 509 |
+
|
| 510 |
+
metrics = {"standard_noise_level": STD, "n_test": N, "exact_match_rate": exact_rate,
|
| 511 |
+
"count_mae": count_mae, "occupancy_mae": occ_mae,
|
| 512 |
+
"noise_sweep_levels": levels, "noise_sweep_exact_match_rates": sweep,
|
| 513 |
+
"method": "Classical CV: color-distance segmentation -> morphology -> connected "
|
| 514 |
+
"components -> geometric filtering. No pretrained weights."}
|
| 515 |
+
with open(os.path.join(DATA_DIR, "vision_eval.json"), "w") as f:
|
| 516 |
+
json.dump(metrics, f, indent=2)
|
| 517 |
+
return metrics
|
| 518 |
+
|
| 519 |
+
|
| 520 |
if __name__ == "__main__":
|
| 521 |
intent_metrics = build_intent_classifier()
|
| 522 |
anomaly_metrics = build_anomaly_detector()
|
| 523 |
build_retrieval_eval()
|
| 524 |
build_inventory_and_orders()
|
| 525 |
build_latency_benchmark(intent_metrics, anomaly_metrics)
|
| 526 |
+
build_vision_evaluation()
|
| 527 |
+
build_sample_batch_csv()
|
| 528 |
print("\nAll artifacts built successfully.")
|
data/intent_dataset.csv
CHANGED
|
@@ -1,141 +1,141 @@
|
|
| 1 |
text,intent
|
| 2 |
-
What's the fastest picking route for order #
|
| 3 |
Pause AGV-12 until the aisle in Zone E is clear,agv_navigation
|
| 4 |
-
Is order #
|
| 5 |
Report unsafe pallet stacking in Zone F,safety_incident
|
| 6 |
-
What's the status of order #
|
| 7 |
What's the current uptime percentage for Zone D?,system_status
|
| 8 |
Check system health for the receiving dock,system_status
|
| 9 |
"A spill was reported near Crane-05 in the receiving dock, needs cleanup",safety_incident
|
| 10 |
What's the most efficient picking sequence for the receiving dock today?,picking_optimization
|
| 11 |
Schedule maintenance for Sorter-06,equipment_maintenance
|
| 12 |
Explain how an AS/RS works,general_faq
|
| 13 |
-
When will order #
|
| 14 |
The sorter in Zone A keeps jamming,equipment_maintenance
|
| 15 |
-
Track order #
|
| 16 |
Report unsafe pallet stacking in the shipping dock,safety_incident
|
| 17 |
Why is inventory accuracy important?,general_faq
|
| 18 |
Crane Crane-05 reported a fault code,equipment_maintenance
|
| 19 |
Would zone picking work better than batch picking for the cross-dock area?,picking_optimization
|
| 20 |
-
Is order #
|
| 21 |
-
Has order #
|
| 22 |
-
Do we have enough SKU-
|
| 23 |
Send the next available AGV to picking station 22,agv_navigation
|
| 24 |
-
What is the current stock level for SKU-
|
| 25 |
Check system health for Zone A,system_status
|
| 26 |
-
Give me stock levels across all zones for SKU-
|
| 27 |
Report vibration issue on Sorter-06,equipment_maintenance
|
| 28 |
There was a near collision between AGV-12 and a pedestrian in Zone F,safety_incident
|
| 29 |
Send AGV Sorter-06 to the mezzanine,agv_navigation
|
| 30 |
What does WCS stand for?,general_faq
|
| 31 |
Report vibration issue on Crane-09,equipment_maintenance
|
| 32 |
Is the sorter in the shipping dock running normally?,system_status
|
| 33 |
-
What's the status of order #
|
| 34 |
-
Show me the on-hand quantity of SKU-
|
| 35 |
File an incident report for Zone F,safety_incident
|
| 36 |
A forklift near-miss was reported in the mezzanine,safety_incident
|
| 37 |
Explain the difference between discrete and batch picking,general_faq
|
| 38 |
-
Show the fulfillment status of #
|
| 39 |
"Belt Crane-03 stopped unexpectedly, please check",equipment_maintenance
|
| 40 |
File an incident report for Zone C,safety_incident
|
| 41 |
Is the WCS connection to AGV-12 stable?,system_status
|
| 42 |
-
Why hasn't order #
|
| 43 |
What is the current location of Conveyor-14?,agv_navigation
|
| 44 |
-
What's the fastest picking route for order #
|
| 45 |
What is the uptime for AGV-18 today?,system_status
|
| 46 |
Send AGV Crane-09 to Zone D,agv_navigation
|
| 47 |
-
When will order #
|
| 48 |
-
Has order #
|
| 49 |
-
Give me stock levels across all zones for SKU-
|
| 50 |
"Someone bypassed the safety gate near Crane-05, please log this",safety_incident
|
| 51 |
Is the sorter in the receiving dock running normally?,system_status
|
| 52 |
Is the WCS connection to AGV-18 stable?,system_status
|
| 53 |
Sorter-02 motor temperature seems high,equipment_maintenance
|
| 54 |
-
Is SKU-
|
| 55 |
Is the WCS connection to Crane-05 stable?,system_status
|
| 56 |
-
What is the current stock level for SKU-
|
| 57 |
-
Is order #
|
| 58 |
Redirect AMR-33 around the blocked aisle in the mezzanine,agv_navigation
|
| 59 |
-
What is the current stock level for SKU-
|
| 60 |
-
Do we have enough SKU-
|
| 61 |
-
What's the fastest picking route for order #
|
| 62 |
Log a safety incident involving Conveyor-14,safety_incident
|
| 63 |
-
How many units of SKU-
|
| 64 |
Suggest a wave picking plan for Zone B,picking_optimization
|
| 65 |
-
Track order #
|
| 66 |
Route Sorter-06 to picking station 18,agv_navigation
|
| 67 |
How should we sequence orders to minimize walking distance in Zone D?,picking_optimization
|
| 68 |
What is the uptime for Crane-05 today?,system_status
|
| 69 |
Is Crane-05 operational?,system_status
|
| 70 |
Is the WCS connection to Crane-03 stable?,system_status
|
| 71 |
-
What is the current stock level for SKU-
|
| 72 |
Reassign Conveyor-21 to charging station,agv_navigation
|
| 73 |
Why is AGV-07 stuck near Zone E?,agv_navigation
|
| 74 |
-
Show the fulfillment status of #
|
| 75 |
-
Check inventory count for SKU-
|
| 76 |
How should we sequence orders to minimize walking distance in Zone A?,picking_optimization
|
| 77 |
What is wave picking?,general_faq
|
| 78 |
Schedule maintenance for AGV-12,equipment_maintenance
|
| 79 |
How does zone picking work?,general_faq
|
| 80 |
The sorter in the cross-dock area keeps jamming,equipment_maintenance
|
| 81 |
Suggest a wave picking plan for the cross-dock area,picking_optimization
|
| 82 |
-
Show me the on-hand quantity of SKU-
|
| 83 |
-
How much inventory is left for SKU-
|
| 84 |
-
Has order #
|
| 85 |
What is dock-to-stock time?,general_faq
|
| 86 |
Would zone picking work better than batch picking for the mezzanine?,picking_optimization
|
| 87 |
-
Why hasn't order #
|
| 88 |
-
Do we have enough SKU-
|
| 89 |
Has Sorter-02 reported any faults today?,system_status
|
| 90 |
-
How many units of SKU-
|
| 91 |
-
What is the current stock level for SKU-
|
| 92 |
-
Has order #
|
| 93 |
Schedule maintenance for Conveyor-21,equipment_maintenance
|
| 94 |
Should we batch pick these orders together?,picking_optimization
|
| 95 |
-
How much inventory is left for SKU-
|
| 96 |
-
Has order #
|
| 97 |
Check system health for the mezzanine,system_status
|
| 98 |
What's the current uptime percentage for the mezzanine?,system_status
|
| 99 |
Are all cranes online in Zone C?,system_status
|
| 100 |
-
Track order #
|
| 101 |
"AMR-21 seems to be idle near Zone E, please reroute it",agv_navigation
|
| 102 |
How do warehouses reduce energy use during peak hours?,general_faq
|
| 103 |
There was a near collision between Sorter-02 and a pedestrian in the cross-dock area,safety_incident
|
| 104 |
A forklift near-miss was reported in the receiving dock,safety_incident
|
| 105 |
The conveyor belt in Zone E is making noise,equipment_maintenance
|
| 106 |
Schedule maintenance for Sorter-02,equipment_maintenance
|
| 107 |
-
How many units of SKU-
|
| 108 |
-
Do we have enough SKU-
|
| 109 |
Report vibration issue on AGV-12,equipment_maintenance
|
| 110 |
Reassign Crane-09 to charging station,agv_navigation
|
| 111 |
What's the current uptime percentage for Zone A?,system_status
|
| 112 |
What's the most efficient picking sequence for Zone A today?,picking_optimization
|
| 113 |
-
Show me the on-hand quantity of SKU-
|
| 114 |
What's the current uptime percentage for the shipping dock?,system_status
|
| 115 |
How is a WMS different from a WCS?,general_faq
|
| 116 |
Log a safety incident involving Sorter-02,safety_incident
|
| 117 |
-
Give me stock levels across all zones for SKU-
|
| 118 |
-
Is SKU-
|
| 119 |
Please log a near-miss between a pedestrian and Conveyor-14 in the shipping dock,safety_incident
|
| 120 |
-
Is order #
|
| 121 |
Has Conveyor-14 reported any faults today?,system_status
|
| 122 |
"Belt Crane-09 stopped unexpectedly, please check",equipment_maintenance
|
| 123 |
-
Show the fulfillment status of #
|
| 124 |
-
Has order #
|
| 125 |
-
What's the fastest picking route for order #
|
| 126 |
How do sortation systems decide where to route a parcel?,general_faq
|
| 127 |
-
Show me the on-hand quantity of SKU-
|
| 128 |
Is Sorter-06 operational?,system_status
|
| 129 |
-
What's the fastest picking route for order #
|
| 130 |
Can you dispatch AGV-07 to the shipping dock right away?,agv_navigation
|
| 131 |
-
Is SKU-
|
| 132 |
-
Show me the on-hand quantity of SKU-
|
| 133 |
Is AGV-18 operational?,system_status
|
| 134 |
-
What's the status of order #
|
| 135 |
What is the current location of AMR-33?,agv_navigation
|
| 136 |
What is the uptime for Crane-03 today?,system_status
|
| 137 |
-
Why hasn't order #
|
| 138 |
-
Show the fulfillment status of #
|
| 139 |
Are all cranes online in Zone A?,system_status
|
| 140 |
Is the sorter in the mezzanine running normally?,system_status
|
| 141 |
Log a safety incident involving Sorter-06,safety_incident
|
|
@@ -148,21 +148,21 @@ There was a near collision between AGV-18 and a pedestrian in Zone B,safety_inci
|
|
| 148 |
What is the current location of AGV-07?,agv_navigation
|
| 149 |
"A spill was reported near Conveyor-21 in the receiving dock, needs cleanup",safety_incident
|
| 150 |
"AGV-07 seems to be idle near the receiving dock, please reroute it",agv_navigation
|
| 151 |
-
Why hasn't order #
|
| 152 |
Redirect Conveyor-14 around the blocked aisle in the receiving dock,agv_navigation
|
| 153 |
Crane Sorter-02 reported a fault code,equipment_maintenance
|
| 154 |
Pause AGV-18 until the aisle in Zone A is clear,agv_navigation
|
| 155 |
Conveyor-14 motor temperature seems high,equipment_maintenance
|
| 156 |
How should we sequence orders to minimize walking distance in the shipping dock?,picking_optimization
|
| 157 |
"Sorter-02 seems to be idle near Zone C, please reroute it",agv_navigation
|
| 158 |
-
Has order #
|
| 159 |
Reassign AMR-21 to charging station,agv_navigation
|
| 160 |
"A spill was reported near AMR-21 in the shipping dock, needs cleanup",safety_incident
|
| 161 |
-
Why hasn't order #
|
| 162 |
What is the current location of AMR-21?,agv_navigation
|
| 163 |
Schedule maintenance for AGV-18,equipment_maintenance
|
| 164 |
Reassign AMR-33 to charging station,agv_navigation
|
| 165 |
-
What's the status of order #
|
| 166 |
How should we sequence orders to minimize walking distance in Zone E?,picking_optimization
|
| 167 |
What is the uptime for Conveyor-14 today?,system_status
|
| 168 |
Are all cranes online in Zone F?,system_status
|
|
@@ -172,28 +172,28 @@ Can you dispatch Conveyor-21 to Zone F right away?,agv_navigation
|
|
| 172 |
How can we reduce travel time for pickers in Zone C?,picking_optimization
|
| 173 |
Is the sorter in Zone E running normally?,system_status
|
| 174 |
Report unsafe pallet stacking in Zone E,safety_incident
|
| 175 |
-
What is the current stock level for SKU-
|
| 176 |
The sorter in Zone D keeps jamming,equipment_maintenance
|
| 177 |
"A worker slipped near AGV-18, please log it",safety_incident
|
| 178 |
Check system health for Zone F,system_status
|
| 179 |
-
What's the fastest picking route for order #
|
| 180 |
Would zone picking work better than batch picking for Zone C?,picking_optimization
|
| 181 |
-
How much inventory is left for SKU-
|
| 182 |
-
Track order #
|
| 183 |
Route Crane-05 to picking station 18,agv_navigation
|
| 184 |
Has AGV-12 reported any faults today?,system_status
|
| 185 |
-
Give me stock levels across all zones for SKU-
|
| 186 |
Is AMR-21 operational?,system_status
|
| 187 |
A forklift near-miss was reported in Zone D,safety_incident
|
| 188 |
"Belt Conveyor-21 stopped unexpectedly, please check",equipment_maintenance
|
| 189 |
"Belt Sorter-06 stopped unexpectedly, please check",equipment_maintenance
|
| 190 |
Report vibration issue on Conveyor-14,equipment_maintenance
|
| 191 |
Reassign Crane-05 to charging station,agv_navigation
|
| 192 |
-
Give me stock levels across all zones for SKU-
|
| 193 |
Has AGV-18 reported any faults today?,system_status
|
| 194 |
-
Track order #
|
| 195 |
-
When will order #
|
| 196 |
-
Check inventory count for SKU-
|
| 197 |
What's the most efficient picking sequence for the shipping dock today?,picking_optimization
|
| 198 |
Log a breakdown for Crane-09 in Zone F,equipment_maintenance
|
| 199 |
Has Sorter-06 reported any faults today?,system_status
|
|
@@ -203,34 +203,34 @@ What's the most efficient picking sequence for Zone B today?,picking_optimizatio
|
|
| 203 |
Route Sorter-06 to picking station 3,agv_navigation
|
| 204 |
Is the sorter in Zone D running normally?,system_status
|
| 205 |
What is the current location of Sorter-02?,agv_navigation
|
| 206 |
-
Has order #
|
| 207 |
Please log a near-miss between a pedestrian and Crane-03 in Zone E,safety_incident
|
| 208 |
Log a breakdown for Conveyor-21 in Zone C,equipment_maintenance
|
| 209 |
How should we sequence orders to minimize walking distance in Zone C?,picking_optimization
|
| 210 |
The conveyor belt in Zone B is making noise,equipment_maintenance
|
| 211 |
-
What is the current stock level for SKU-
|
| 212 |
-
When will order #
|
| 213 |
File an incident report for Zone E,safety_incident
|
| 214 |
-
Show the fulfillment status of #
|
| 215 |
Report unsafe pallet stacking in the receiving dock,safety_incident
|
| 216 |
Crane Crane-09 reported a fault code,equipment_maintenance
|
| 217 |
"A worker slipped near AMR-33, please log it",safety_incident
|
| 218 |
-
When will order #
|
| 219 |
Report vibration issue on Crane-03,equipment_maintenance
|
| 220 |
Log a safety incident involving AMR-33,safety_incident
|
| 221 |
Please log a near-miss between a pedestrian and AGV-12 in the shipping dock,safety_incident
|
| 222 |
How can we reduce travel time for pickers in Zone A?,picking_optimization
|
| 223 |
-
Why hasn't order #
|
| 224 |
What's the most efficient picking sequence for Zone E today?,picking_optimization
|
| 225 |
-
Why hasn't order #
|
| 226 |
-
Is SKU-
|
| 227 |
-
Is SKU-
|
| 228 |
The conveyor belt in Zone A is making noise,equipment_maintenance
|
| 229 |
Crane-09 motor temperature seems high,equipment_maintenance
|
| 230 |
Suggest a wave picking plan for Zone D,picking_optimization
|
| 231 |
Reassign Crane-03 to charging station,agv_navigation
|
| 232 |
Send the next available AGV to picking station 12,agv_navigation
|
| 233 |
-
Is order #
|
| 234 |
Can you dispatch AGV-12 to the receiving dock right away?,agv_navigation
|
| 235 |
Is Conveyor-14 operational?,system_status
|
| 236 |
Would zone picking work better than batch picking for the shipping dock?,picking_optimization
|
|
@@ -241,10 +241,10 @@ What's the most efficient picking sequence for Zone F today?,picking_optimizatio
|
|
| 241 |
Crane AGV-18 reported a fault code,equipment_maintenance
|
| 242 |
AMR-21 motor temperature seems high,equipment_maintenance
|
| 243 |
Please log a near-miss between a pedestrian and Crane-09 in the mezzanine,safety_incident
|
| 244 |
-
What's the fastest picking route for order #
|
| 245 |
Reassign AGV-12 to charging station,agv_navigation
|
| 246 |
"Conveyor-21 seems to be idle near Zone B, please reroute it",agv_navigation
|
| 247 |
-
When will order #
|
| 248 |
Is Sorter-02 operational?,system_status
|
| 249 |
Pause Crane-05 until the aisle in the receiving dock is clear,agv_navigation
|
| 250 |
Crane Conveyor-21 reported a fault code,equipment_maintenance
|
|
@@ -257,94 +257,94 @@ Log a safety incident involving Conveyor-21,safety_incident
|
|
| 257 |
Please log a near-miss between a pedestrian and Crane-09 in Zone A,safety_incident
|
| 258 |
What's the current uptime percentage for the cross-dock area?,system_status
|
| 259 |
There was a near collision between Crane-05 and a pedestrian in Zone F,safety_incident
|
| 260 |
-
Has order #
|
| 261 |
Log a breakdown for Sorter-02 in Zone B,equipment_maintenance
|
| 262 |
Recommend a picking strategy for high-velocity SKUs,picking_optimization
|
| 263 |
-
What's the fastest picking route for order #
|
| 264 |
How should we sequence orders to minimize walking distance in Zone F?,picking_optimization
|
| 265 |
What's the difference between AGV and AMR?,general_faq
|
| 266 |
"A worker slipped near Crane-03, please log it",safety_incident
|
| 267 |
"AGV-18 seems to be idle near Zone E, please reroute it",agv_navigation
|
| 268 |
-
What's the fastest picking route for order #
|
| 269 |
The conveyor belt in the mezzanine is making noise,equipment_maintenance
|
| 270 |
"Belt AMR-21 stopped unexpectedly, please check",equipment_maintenance
|
| 271 |
Redirect AMR-33 around the blocked aisle in Zone F,agv_navigation
|
| 272 |
Would zone picking work better than batch picking for Zone A?,picking_optimization
|
| 273 |
Why is Conveyor-14 stuck near the receiving dock?,agv_navigation
|
| 274 |
-
Has order #
|
| 275 |
Log a breakdown for Crane-05 in Zone F,equipment_maintenance
|
| 276 |
Why is AGV-07 stuck near Zone B?,agv_navigation
|
| 277 |
There was a near collision between Conveyor-21 and a pedestrian in Zone D,safety_incident
|
| 278 |
-
What's the fastest picking route for order #
|
| 279 |
"Belt AMR-33 stopped unexpectedly, please check",equipment_maintenance
|
| 280 |
Is the sorter in Zone C running normally?,system_status
|
| 281 |
-
Is order #
|
| 282 |
Check system health for the shipping dock,system_status
|
| 283 |
Crane Conveyor-14 reported a fault code,equipment_maintenance
|
| 284 |
There was a near collision between AGV-12 and a pedestrian in Zone D,safety_incident
|
| 285 |
-
Is SKU-
|
| 286 |
Check system health for Zone D,system_status
|
| 287 |
-
Has order #
|
| 288 |
Would zone picking work better than batch picking for Zone D?,picking_optimization
|
| 289 |
-
When will order #
|
| 290 |
Suggest a wave picking plan for Zone C,picking_optimization
|
| 291 |
Can you dispatch AGV-07 to Zone F right away?,agv_navigation
|
| 292 |
The conveyor belt in the cross-dock area is making noise,equipment_maintenance
|
| 293 |
-
How many units of SKU-
|
| 294 |
-
Show me the on-hand quantity of SKU-
|
| 295 |
-
Check inventory count for SKU-
|
| 296 |
Crane Crane-03 reported a fault code,equipment_maintenance
|
| 297 |
Schedule maintenance for Crane-05,equipment_maintenance
|
| 298 |
Report unsafe pallet stacking in Zone D,safety_incident
|
| 299 |
"Someone bypassed the safety gate near Crane-09, please log this",safety_incident
|
| 300 |
"AMR-21 seems to be idle near Zone B, please reroute it",agv_navigation
|
| 301 |
-
Why hasn't order #
|
| 302 |
-
Is SKU-
|
| 303 |
Would zone picking work better than batch picking for the receiving dock?,picking_optimization
|
| 304 |
What is a mini-load system?,general_faq
|
| 305 |
"A spill was reported near Conveyor-14 in the receiving dock, needs cleanup",safety_incident
|
| 306 |
Report vibration issue on Sorter-02,equipment_maintenance
|
| 307 |
-
Do we have enough SKU-
|
| 308 |
Why do warehouses use cycle counting instead of annual counts?,general_faq
|
| 309 |
Route Crane-05 to picking station 12,agv_navigation
|
| 310 |
-
What's the fastest picking route for order #
|
| 311 |
Suggest a wave picking plan for the mezzanine,picking_optimization
|
| 312 |
Is the WCS connection to AGV-07 stable?,system_status
|
| 313 |
-
Do we have enough SKU-
|
| 314 |
How should we sequence orders to minimize walking distance in the cross-dock area?,picking_optimization
|
| 315 |
Is the sorter in the cross-dock area running normally?,system_status
|
| 316 |
-
Track order #
|
| 317 |
Pause Sorter-06 until the aisle in Zone E is clear,agv_navigation
|
| 318 |
Check system health for Zone B,system_status
|
| 319 |
Pause Crane-03 until the aisle in the cross-dock area is clear,agv_navigation
|
| 320 |
The sorter in Zone F keeps jamming,equipment_maintenance
|
| 321 |
-
What's the status of order #
|
| 322 |
Optimize the pick path for Zone F,picking_optimization
|
| 323 |
Send AGV AMR-21 to Zone D,agv_navigation
|
| 324 |
-
Show the fulfillment status of #
|
| 325 |
-
Check inventory count for SKU-
|
| 326 |
-
What is the current stock level for SKU-
|
| 327 |
What's the most efficient picking sequence for the cross-dock area today?,picking_optimization
|
| 328 |
"AMR-33 seems to be idle near Zone E, please reroute it",agv_navigation
|
| 329 |
Has AMR-33 reported any faults today?,system_status
|
| 330 |
-
Give me stock levels across all zones for SKU-
|
| 331 |
What is the uptime for AMR-33 today?,system_status
|
| 332 |
Route Crane-03 to picking station 18,agv_navigation
|
| 333 |
How can we reduce travel time for pickers in Zone F?,picking_optimization
|
| 334 |
-
What's the fastest picking route for order #
|
| 335 |
Report vibration issue on AMR-33,equipment_maintenance
|
| 336 |
Check system health for Zone E,system_status
|
| 337 |
"Belt AGV-07 stopped unexpectedly, please check",equipment_maintenance
|
| 338 |
-
Is order #
|
| 339 |
The sorter in the shipping dock keeps jamming,equipment_maintenance
|
| 340 |
Report vibration issue on Conveyor-21,equipment_maintenance
|
| 341 |
-
Give me stock levels across all zones for SKU-
|
| 342 |
"A spill was reported near AMR-21 in the cross-dock area, needs cleanup",safety_incident
|
| 343 |
-
Show the fulfillment status of #
|
| 344 |
Please log a near-miss between a pedestrian and Conveyor-21 in the receiving dock,safety_incident
|
| 345 |
-
Is SKU-
|
| 346 |
"Belt Conveyor-14 stopped unexpectedly, please check",equipment_maintenance
|
| 347 |
-
What is the current stock level for SKU-
|
| 348 |
Are all cranes online in Zone B?,system_status
|
| 349 |
"A worker slipped near AMR-21, please log it",safety_incident
|
| 350 |
What is the current location of AGV-18?,agv_navigation
|
|
@@ -364,30 +364,30 @@ There was a near collision between AMR-33 and a pedestrian in Zone C,safety_inci
|
|
| 364 |
Would zone picking work better than batch picking for Zone E?,picking_optimization
|
| 365 |
Send the next available AGV to picking station 3,agv_navigation
|
| 366 |
Crane-03 motor temperature seems high,equipment_maintenance
|
| 367 |
-
Is SKU-
|
| 368 |
Why is AGV-12 stuck near the shipping dock?,agv_navigation
|
| 369 |
Send the next available AGV to picking station 5,agv_navigation
|
| 370 |
Please log a near-miss between a pedestrian and AGV-18 in Zone C,safety_incident
|
| 371 |
What's the most efficient picking sequence for Zone C today?,picking_optimization
|
| 372 |
-
Track order #
|
| 373 |
-
Is SKU-
|
| 374 |
-
Do we have enough SKU-
|
| 375 |
How does regenerative braking work on an AS/RS crane?,general_faq
|
| 376 |
-
Give me stock levels across all zones for SKU-
|
| 377 |
-
When will order #
|
| 378 |
Why is AGV-12 stuck near the receiving dock?,agv_navigation
|
| 379 |
-
Give me stock levels across all zones for SKU-
|
| 380 |
Optimize the pick path for Zone D,picking_optimization
|
| 381 |
Give me the current status of the WMS integration,system_status
|
| 382 |
-
When will order #
|
| 383 |
A forklift near-miss was reported in Zone F,safety_incident
|
| 384 |
-
What is the current stock level for SKU-
|
| 385 |
"Someone bypassed the safety gate near AMR-21, please log this",safety_incident
|
| 386 |
How can we reduce travel time for pickers in Zone D?,picking_optimization
|
| 387 |
How can we reduce travel time for pickers in the shipping dock?,picking_optimization
|
| 388 |
-
Give me stock levels across all zones for SKU-
|
| 389 |
"A spill was reported near AGV-12 in Zone F, needs cleanup",safety_incident
|
| 390 |
-
Track order #
|
| 391 |
Are all cranes online in Zone E?,system_status
|
| 392 |
Would zone picking work better than batch picking for Zone B?,picking_optimization
|
| 393 |
The sorter in the mezzanine keeps jamming,equipment_maintenance
|
|
@@ -397,24 +397,24 @@ What is a WMS?,general_faq
|
|
| 397 |
A forklift near-miss was reported in Zone A,safety_incident
|
| 398 |
What is predictive maintenance?,general_faq
|
| 399 |
There was a near collision between Conveyor-14 and a pedestrian in Zone B,safety_incident
|
| 400 |
-
What's the fastest picking route for order #
|
| 401 |
-
Do we have enough SKU-
|
| 402 |
Are all cranes online in the mezzanine?,system_status
|
| 403 |
-
Do we have enough SKU-
|
| 404 |
Crane AMR-33 reported a fault code,equipment_maintenance
|
| 405 |
Pause Conveyor-14 until the aisle in Zone E is clear,agv_navigation
|
| 406 |
What's the difference between a stacker crane and a shuttle?,general_faq
|
| 407 |
Redirect Sorter-06 around the blocked aisle in Zone C,agv_navigation
|
| 408 |
Conveyor-21 motor temperature seems high,equipment_maintenance
|
| 409 |
-
Do we have enough SKU-
|
| 410 |
-
Check inventory count for SKU-
|
| 411 |
Report unsafe pallet stacking in Zone C,safety_incident
|
| 412 |
Redirect Crane-09 around the blocked aisle in the mezzanine,agv_navigation
|
| 413 |
How should we sequence orders to minimize walking distance in the mezzanine?,picking_optimization
|
| 414 |
-
Do we have enough SKU-
|
| 415 |
Please log a near-miss between a pedestrian and Crane-05 in the mezzanine,safety_incident
|
| 416 |
The conveyor belt in Zone F is making noise,equipment_maintenance
|
| 417 |
-
What's the fastest picking route for order #
|
| 418 |
Can you dispatch Conveyor-14 to the receiving dock right away?,agv_navigation
|
| 419 |
What's the current uptime percentage for the receiving dock?,system_status
|
| 420 |
Optimize the pick path for Zone B,picking_optimization
|
|
@@ -426,18 +426,18 @@ Log a breakdown for Conveyor-14 in the shipping dock,equipment_maintenance
|
|
| 426 |
The conveyor belt in the shipping dock is making noise,equipment_maintenance
|
| 427 |
Log a breakdown for Crane-05 in the receiving dock,equipment_maintenance
|
| 428 |
There was a near collision between Sorter-06 and a pedestrian in Zone E,safety_incident
|
| 429 |
-
What's the status of order #
|
| 430 |
What's the current uptime percentage for Zone B?,system_status
|
| 431 |
-
Track order #
|
| 432 |
"A spill was reported near AGV-07 in the receiving dock, needs cleanup",safety_incident
|
| 433 |
Pause AGV-07 until the aisle in the cross-dock area is clear,agv_navigation
|
| 434 |
AGV-07 motor temperature seems high,equipment_maintenance
|
| 435 |
What KPIs matter most in warehouse automation?,general_faq
|
| 436 |
-
What is the current stock level for SKU-
|
| 437 |
Redirect Crane-09 around the blocked aisle in Zone E,agv_navigation
|
| 438 |
-
Is order #
|
| 439 |
-
Has order #
|
| 440 |
-
What's the fastest picking route for order #
|
| 441 |
There was a near collision between AGV-12 and a pedestrian in Zone C,safety_incident
|
| 442 |
What is a pick rate and how is it measured?,general_faq
|
| 443 |
-
Why hasn't order #
|
|
|
|
| 1 |
text,intent
|
| 2 |
+
What's the fastest picking route for order #85330?,picking_optimization
|
| 3 |
Pause AGV-12 until the aisle in Zone E is clear,agv_navigation
|
| 4 |
+
Is order #82421 delayed?,order_status
|
| 5 |
Report unsafe pallet stacking in Zone F,safety_incident
|
| 6 |
+
What's the status of order #47013?,order_status
|
| 7 |
What's the current uptime percentage for Zone D?,system_status
|
| 8 |
Check system health for the receiving dock,system_status
|
| 9 |
"A spill was reported near Crane-05 in the receiving dock, needs cleanup",safety_incident
|
| 10 |
What's the most efficient picking sequence for the receiving dock today?,picking_optimization
|
| 11 |
Schedule maintenance for Sorter-06,equipment_maintenance
|
| 12 |
Explain how an AS/RS works,general_faq
|
| 13 |
+
When will order #62450 be delivered?,order_status
|
| 14 |
The sorter in Zone A keeps jamming,equipment_maintenance
|
| 15 |
+
Track order #33029 for me,order_status
|
| 16 |
Report unsafe pallet stacking in the shipping dock,safety_incident
|
| 17 |
Why is inventory accuracy important?,general_faq
|
| 18 |
Crane Crane-05 reported a fault code,equipment_maintenance
|
| 19 |
Would zone picking work better than batch picking for the cross-dock area?,picking_optimization
|
| 20 |
+
Is order #15110 delayed?,order_status
|
| 21 |
+
Has order #43202 shipped yet?,order_status
|
| 22 |
+
Do we have enough SKU-2103 to fulfill 200 units?,inventory_check
|
| 23 |
Send the next available AGV to picking station 22,agv_navigation
|
| 24 |
+
What is the current stock level for SKU-8359?,inventory_check
|
| 25 |
Check system health for Zone A,system_status
|
| 26 |
+
Give me stock levels across all zones for SKU-9769,inventory_check
|
| 27 |
Report vibration issue on Sorter-06,equipment_maintenance
|
| 28 |
There was a near collision between AGV-12 and a pedestrian in Zone F,safety_incident
|
| 29 |
Send AGV Sorter-06 to the mezzanine,agv_navigation
|
| 30 |
What does WCS stand for?,general_faq
|
| 31 |
Report vibration issue on Crane-09,equipment_maintenance
|
| 32 |
Is the sorter in the shipping dock running normally?,system_status
|
| 33 |
+
What's the status of order #32613?,order_status
|
| 34 |
+
Show me the on-hand quantity of SKU-1747,inventory_check
|
| 35 |
File an incident report for Zone F,safety_incident
|
| 36 |
A forklift near-miss was reported in the mezzanine,safety_incident
|
| 37 |
Explain the difference between discrete and batch picking,general_faq
|
| 38 |
+
Show the fulfillment status of #41155,order_status
|
| 39 |
"Belt Crane-03 stopped unexpectedly, please check",equipment_maintenance
|
| 40 |
File an incident report for Zone C,safety_incident
|
| 41 |
Is the WCS connection to AGV-12 stable?,system_status
|
| 42 |
+
Why hasn't order #69979 left the dock yet?,order_status
|
| 43 |
What is the current location of Conveyor-14?,agv_navigation
|
| 44 |
+
What's the fastest picking route for order #25183?,picking_optimization
|
| 45 |
What is the uptime for AGV-18 today?,system_status
|
| 46 |
Send AGV Crane-09 to Zone D,agv_navigation
|
| 47 |
+
When will order #87283 be delivered?,order_status
|
| 48 |
+
Has order #57411 shipped yet?,order_status
|
| 49 |
+
Give me stock levels across all zones for SKU-5185,inventory_check
|
| 50 |
"Someone bypassed the safety gate near Crane-05, please log this",safety_incident
|
| 51 |
Is the sorter in the receiving dock running normally?,system_status
|
| 52 |
Is the WCS connection to AGV-18 stable?,system_status
|
| 53 |
Sorter-02 motor temperature seems high,equipment_maintenance
|
| 54 |
+
Is SKU-1506 in stock at the cross-dock area?,inventory_check
|
| 55 |
Is the WCS connection to Crane-05 stable?,system_status
|
| 56 |
+
What is the current stock level for SKU-9304?,inventory_check
|
| 57 |
+
Is order #61347 delayed?,order_status
|
| 58 |
Redirect AMR-33 around the blocked aisle in the mezzanine,agv_navigation
|
| 59 |
+
What is the current stock level for SKU-5753?,inventory_check
|
| 60 |
+
Do we have enough SKU-1627 to fulfill 200 units?,inventory_check
|
| 61 |
+
What's the fastest picking route for order #75573?,picking_optimization
|
| 62 |
Log a safety incident involving Conveyor-14,safety_incident
|
| 63 |
+
How many units of SKU-9661 are in Zone F?,inventory_check
|
| 64 |
Suggest a wave picking plan for Zone B,picking_optimization
|
| 65 |
+
Track order #66375 for me,order_status
|
| 66 |
Route Sorter-06 to picking station 18,agv_navigation
|
| 67 |
How should we sequence orders to minimize walking distance in Zone D?,picking_optimization
|
| 68 |
What is the uptime for Crane-05 today?,system_status
|
| 69 |
Is Crane-05 operational?,system_status
|
| 70 |
Is the WCS connection to Crane-03 stable?,system_status
|
| 71 |
+
What is the current stock level for SKU-5292?,inventory_check
|
| 72 |
Reassign Conveyor-21 to charging station,agv_navigation
|
| 73 |
Why is AGV-07 stuck near Zone E?,agv_navigation
|
| 74 |
+
Show the fulfillment status of #50488,order_status
|
| 75 |
+
Check inventory count for SKU-4502 in the shipping dock,inventory_check
|
| 76 |
How should we sequence orders to minimize walking distance in Zone A?,picking_optimization
|
| 77 |
What is wave picking?,general_faq
|
| 78 |
Schedule maintenance for AGV-12,equipment_maintenance
|
| 79 |
How does zone picking work?,general_faq
|
| 80 |
The sorter in the cross-dock area keeps jamming,equipment_maintenance
|
| 81 |
Suggest a wave picking plan for the cross-dock area,picking_optimization
|
| 82 |
+
Show me the on-hand quantity of SKU-3554,inventory_check
|
| 83 |
+
How much inventory is left for SKU-4486?,inventory_check
|
| 84 |
+
Has order #21753 shipped yet?,order_status
|
| 85 |
What is dock-to-stock time?,general_faq
|
| 86 |
Would zone picking work better than batch picking for the mezzanine?,picking_optimization
|
| 87 |
+
Why hasn't order #32297 left the dock yet?,order_status
|
| 88 |
+
Do we have enough SKU-6904 to fulfill 200 units?,inventory_check
|
| 89 |
Has Sorter-02 reported any faults today?,system_status
|
| 90 |
+
How many units of SKU-5094 are in Zone F?,inventory_check
|
| 91 |
+
What is the current stock level for SKU-1726?,inventory_check
|
| 92 |
+
Has order #92929 shipped yet?,order_status
|
| 93 |
Schedule maintenance for Conveyor-21,equipment_maintenance
|
| 94 |
Should we batch pick these orders together?,picking_optimization
|
| 95 |
+
How much inventory is left for SKU-9428?,inventory_check
|
| 96 |
+
Has order #65652 shipped yet?,order_status
|
| 97 |
Check system health for the mezzanine,system_status
|
| 98 |
What's the current uptime percentage for the mezzanine?,system_status
|
| 99 |
Are all cranes online in Zone C?,system_status
|
| 100 |
+
Track order #70218 for me,order_status
|
| 101 |
"AMR-21 seems to be idle near Zone E, please reroute it",agv_navigation
|
| 102 |
How do warehouses reduce energy use during peak hours?,general_faq
|
| 103 |
There was a near collision between Sorter-02 and a pedestrian in the cross-dock area,safety_incident
|
| 104 |
A forklift near-miss was reported in the receiving dock,safety_incident
|
| 105 |
The conveyor belt in Zone E is making noise,equipment_maintenance
|
| 106 |
Schedule maintenance for Sorter-02,equipment_maintenance
|
| 107 |
+
How many units of SKU-6575 are in the shipping dock?,inventory_check
|
| 108 |
+
Do we have enough SKU-5018 to fulfill 200 units?,inventory_check
|
| 109 |
Report vibration issue on AGV-12,equipment_maintenance
|
| 110 |
Reassign Crane-09 to charging station,agv_navigation
|
| 111 |
What's the current uptime percentage for Zone A?,system_status
|
| 112 |
What's the most efficient picking sequence for Zone A today?,picking_optimization
|
| 113 |
+
Show me the on-hand quantity of SKU-9298,inventory_check
|
| 114 |
What's the current uptime percentage for the shipping dock?,system_status
|
| 115 |
How is a WMS different from a WCS?,general_faq
|
| 116 |
Log a safety incident involving Sorter-02,safety_incident
|
| 117 |
+
Give me stock levels across all zones for SKU-1808,inventory_check
|
| 118 |
+
Is SKU-9799 in stock at Zone C?,inventory_check
|
| 119 |
Please log a near-miss between a pedestrian and Conveyor-14 in the shipping dock,safety_incident
|
| 120 |
+
Is order #20803 delayed?,order_status
|
| 121 |
Has Conveyor-14 reported any faults today?,system_status
|
| 122 |
"Belt Crane-09 stopped unexpectedly, please check",equipment_maintenance
|
| 123 |
+
Show the fulfillment status of #65677,order_status
|
| 124 |
+
Has order #79386 shipped yet?,order_status
|
| 125 |
+
What's the fastest picking route for order #91540?,picking_optimization
|
| 126 |
How do sortation systems decide where to route a parcel?,general_faq
|
| 127 |
+
Show me the on-hand quantity of SKU-4036,inventory_check
|
| 128 |
Is Sorter-06 operational?,system_status
|
| 129 |
+
What's the fastest picking route for order #65982?,picking_optimization
|
| 130 |
Can you dispatch AGV-07 to the shipping dock right away?,agv_navigation
|
| 131 |
+
Is SKU-8044 in stock at Zone D?,inventory_check
|
| 132 |
+
Show me the on-hand quantity of SKU-8274,inventory_check
|
| 133 |
Is AGV-18 operational?,system_status
|
| 134 |
+
What's the status of order #12781?,order_status
|
| 135 |
What is the current location of AMR-33?,agv_navigation
|
| 136 |
What is the uptime for Crane-03 today?,system_status
|
| 137 |
+
Why hasn't order #62037 left the dock yet?,order_status
|
| 138 |
+
Show the fulfillment status of #22436,order_status
|
| 139 |
Are all cranes online in Zone A?,system_status
|
| 140 |
Is the sorter in the mezzanine running normally?,system_status
|
| 141 |
Log a safety incident involving Sorter-06,safety_incident
|
|
|
|
| 148 |
What is the current location of AGV-07?,agv_navigation
|
| 149 |
"A spill was reported near Conveyor-21 in the receiving dock, needs cleanup",safety_incident
|
| 150 |
"AGV-07 seems to be idle near the receiving dock, please reroute it",agv_navigation
|
| 151 |
+
Why hasn't order #63053 left the dock yet?,order_status
|
| 152 |
Redirect Conveyor-14 around the blocked aisle in the receiving dock,agv_navigation
|
| 153 |
Crane Sorter-02 reported a fault code,equipment_maintenance
|
| 154 |
Pause AGV-18 until the aisle in Zone A is clear,agv_navigation
|
| 155 |
Conveyor-14 motor temperature seems high,equipment_maintenance
|
| 156 |
How should we sequence orders to minimize walking distance in the shipping dock?,picking_optimization
|
| 157 |
"Sorter-02 seems to be idle near Zone C, please reroute it",agv_navigation
|
| 158 |
+
Has order #91932 shipped yet?,order_status
|
| 159 |
Reassign AMR-21 to charging station,agv_navigation
|
| 160 |
"A spill was reported near AMR-21 in the shipping dock, needs cleanup",safety_incident
|
| 161 |
+
Why hasn't order #11052 left the dock yet?,order_status
|
| 162 |
What is the current location of AMR-21?,agv_navigation
|
| 163 |
Schedule maintenance for AGV-18,equipment_maintenance
|
| 164 |
Reassign AMR-33 to charging station,agv_navigation
|
| 165 |
+
What's the status of order #47552?,order_status
|
| 166 |
How should we sequence orders to minimize walking distance in Zone E?,picking_optimization
|
| 167 |
What is the uptime for Conveyor-14 today?,system_status
|
| 168 |
Are all cranes online in Zone F?,system_status
|
|
|
|
| 172 |
How can we reduce travel time for pickers in Zone C?,picking_optimization
|
| 173 |
Is the sorter in Zone E running normally?,system_status
|
| 174 |
Report unsafe pallet stacking in Zone E,safety_incident
|
| 175 |
+
What is the current stock level for SKU-8069?,inventory_check
|
| 176 |
The sorter in Zone D keeps jamming,equipment_maintenance
|
| 177 |
"A worker slipped near AGV-18, please log it",safety_incident
|
| 178 |
Check system health for Zone F,system_status
|
| 179 |
+
What's the fastest picking route for order #91268?,picking_optimization
|
| 180 |
Would zone picking work better than batch picking for Zone C?,picking_optimization
|
| 181 |
+
How much inventory is left for SKU-7962?,inventory_check
|
| 182 |
+
Track order #56354 for me,order_status
|
| 183 |
Route Crane-05 to picking station 18,agv_navigation
|
| 184 |
Has AGV-12 reported any faults today?,system_status
|
| 185 |
+
Give me stock levels across all zones for SKU-9600,inventory_check
|
| 186 |
Is AMR-21 operational?,system_status
|
| 187 |
A forklift near-miss was reported in Zone D,safety_incident
|
| 188 |
"Belt Conveyor-21 stopped unexpectedly, please check",equipment_maintenance
|
| 189 |
"Belt Sorter-06 stopped unexpectedly, please check",equipment_maintenance
|
| 190 |
Report vibration issue on Conveyor-14,equipment_maintenance
|
| 191 |
Reassign Crane-05 to charging station,agv_navigation
|
| 192 |
+
Give me stock levels across all zones for SKU-6051,inventory_check
|
| 193 |
Has AGV-18 reported any faults today?,system_status
|
| 194 |
+
Track order #29623 for me,order_status
|
| 195 |
+
When will order #71699 be delivered?,order_status
|
| 196 |
+
Check inventory count for SKU-5885 in the receiving dock,inventory_check
|
| 197 |
What's the most efficient picking sequence for the shipping dock today?,picking_optimization
|
| 198 |
Log a breakdown for Crane-09 in Zone F,equipment_maintenance
|
| 199 |
Has Sorter-06 reported any faults today?,system_status
|
|
|
|
| 203 |
Route Sorter-06 to picking station 3,agv_navigation
|
| 204 |
Is the sorter in Zone D running normally?,system_status
|
| 205 |
What is the current location of Sorter-02?,agv_navigation
|
| 206 |
+
Has order #18807 shipped yet?,order_status
|
| 207 |
Please log a near-miss between a pedestrian and Crane-03 in Zone E,safety_incident
|
| 208 |
Log a breakdown for Conveyor-21 in Zone C,equipment_maintenance
|
| 209 |
How should we sequence orders to minimize walking distance in Zone C?,picking_optimization
|
| 210 |
The conveyor belt in Zone B is making noise,equipment_maintenance
|
| 211 |
+
What is the current stock level for SKU-5582?,inventory_check
|
| 212 |
+
When will order #26928 be delivered?,order_status
|
| 213 |
File an incident report for Zone E,safety_incident
|
| 214 |
+
Show the fulfillment status of #41446,order_status
|
| 215 |
Report unsafe pallet stacking in the receiving dock,safety_incident
|
| 216 |
Crane Crane-09 reported a fault code,equipment_maintenance
|
| 217 |
"A worker slipped near AMR-33, please log it",safety_incident
|
| 218 |
+
When will order #78867 be delivered?,order_status
|
| 219 |
Report vibration issue on Crane-03,equipment_maintenance
|
| 220 |
Log a safety incident involving AMR-33,safety_incident
|
| 221 |
Please log a near-miss between a pedestrian and AGV-12 in the shipping dock,safety_incident
|
| 222 |
How can we reduce travel time for pickers in Zone A?,picking_optimization
|
| 223 |
+
Why hasn't order #88111 left the dock yet?,order_status
|
| 224 |
What's the most efficient picking sequence for Zone E today?,picking_optimization
|
| 225 |
+
Why hasn't order #13010 left the dock yet?,order_status
|
| 226 |
+
Is SKU-7869 in stock at Zone A?,inventory_check
|
| 227 |
+
Is SKU-5472 in stock at Zone B?,inventory_check
|
| 228 |
The conveyor belt in Zone A is making noise,equipment_maintenance
|
| 229 |
Crane-09 motor temperature seems high,equipment_maintenance
|
| 230 |
Suggest a wave picking plan for Zone D,picking_optimization
|
| 231 |
Reassign Crane-03 to charging station,agv_navigation
|
| 232 |
Send the next available AGV to picking station 12,agv_navigation
|
| 233 |
+
Is order #15312 delayed?,order_status
|
| 234 |
Can you dispatch AGV-12 to the receiving dock right away?,agv_navigation
|
| 235 |
Is Conveyor-14 operational?,system_status
|
| 236 |
Would zone picking work better than batch picking for the shipping dock?,picking_optimization
|
|
|
|
| 241 |
Crane AGV-18 reported a fault code,equipment_maintenance
|
| 242 |
AMR-21 motor temperature seems high,equipment_maintenance
|
| 243 |
Please log a near-miss between a pedestrian and Crane-09 in the mezzanine,safety_incident
|
| 244 |
+
What's the fastest picking route for order #59267?,picking_optimization
|
| 245 |
Reassign AGV-12 to charging station,agv_navigation
|
| 246 |
"Conveyor-21 seems to be idle near Zone B, please reroute it",agv_navigation
|
| 247 |
+
When will order #30267 be delivered?,order_status
|
| 248 |
Is Sorter-02 operational?,system_status
|
| 249 |
Pause Crane-05 until the aisle in the receiving dock is clear,agv_navigation
|
| 250 |
Crane Conveyor-21 reported a fault code,equipment_maintenance
|
|
|
|
| 257 |
Please log a near-miss between a pedestrian and Crane-09 in Zone A,safety_incident
|
| 258 |
What's the current uptime percentage for the cross-dock area?,system_status
|
| 259 |
There was a near collision between Crane-05 and a pedestrian in Zone F,safety_incident
|
| 260 |
+
Has order #89670 shipped yet?,order_status
|
| 261 |
Log a breakdown for Sorter-02 in Zone B,equipment_maintenance
|
| 262 |
Recommend a picking strategy for high-velocity SKUs,picking_optimization
|
| 263 |
+
What's the fastest picking route for order #69155?,picking_optimization
|
| 264 |
How should we sequence orders to minimize walking distance in Zone F?,picking_optimization
|
| 265 |
What's the difference between AGV and AMR?,general_faq
|
| 266 |
"A worker slipped near Crane-03, please log it",safety_incident
|
| 267 |
"AGV-18 seems to be idle near Zone E, please reroute it",agv_navigation
|
| 268 |
+
What's the fastest picking route for order #10844?,picking_optimization
|
| 269 |
The conveyor belt in the mezzanine is making noise,equipment_maintenance
|
| 270 |
"Belt AMR-21 stopped unexpectedly, please check",equipment_maintenance
|
| 271 |
Redirect AMR-33 around the blocked aisle in Zone F,agv_navigation
|
| 272 |
Would zone picking work better than batch picking for Zone A?,picking_optimization
|
| 273 |
Why is Conveyor-14 stuck near the receiving dock?,agv_navigation
|
| 274 |
+
Has order #57746 shipped yet?,order_status
|
| 275 |
Log a breakdown for Crane-05 in Zone F,equipment_maintenance
|
| 276 |
Why is AGV-07 stuck near Zone B?,agv_navigation
|
| 277 |
There was a near collision between Conveyor-21 and a pedestrian in Zone D,safety_incident
|
| 278 |
+
What's the fastest picking route for order #69032?,picking_optimization
|
| 279 |
"Belt AMR-33 stopped unexpectedly, please check",equipment_maintenance
|
| 280 |
Is the sorter in Zone C running normally?,system_status
|
| 281 |
+
Is order #42243 delayed?,order_status
|
| 282 |
Check system health for the shipping dock,system_status
|
| 283 |
Crane Conveyor-14 reported a fault code,equipment_maintenance
|
| 284 |
There was a near collision between AGV-12 and a pedestrian in Zone D,safety_incident
|
| 285 |
+
Is SKU-4734 in stock at Zone E?,inventory_check
|
| 286 |
Check system health for Zone D,system_status
|
| 287 |
+
Has order #76270 shipped yet?,order_status
|
| 288 |
Would zone picking work better than batch picking for Zone D?,picking_optimization
|
| 289 |
+
When will order #32758 be delivered?,order_status
|
| 290 |
Suggest a wave picking plan for Zone C,picking_optimization
|
| 291 |
Can you dispatch AGV-07 to Zone F right away?,agv_navigation
|
| 292 |
The conveyor belt in the cross-dock area is making noise,equipment_maintenance
|
| 293 |
+
How many units of SKU-3646 are in the receiving dock?,inventory_check
|
| 294 |
+
Show me the on-hand quantity of SKU-1123,inventory_check
|
| 295 |
+
Check inventory count for SKU-9034 in Zone D,inventory_check
|
| 296 |
Crane Crane-03 reported a fault code,equipment_maintenance
|
| 297 |
Schedule maintenance for Crane-05,equipment_maintenance
|
| 298 |
Report unsafe pallet stacking in Zone D,safety_incident
|
| 299 |
"Someone bypassed the safety gate near Crane-09, please log this",safety_incident
|
| 300 |
"AMR-21 seems to be idle near Zone B, please reroute it",agv_navigation
|
| 301 |
+
Why hasn't order #86364 left the dock yet?,order_status
|
| 302 |
+
Is SKU-7788 in stock at the receiving dock?,inventory_check
|
| 303 |
Would zone picking work better than batch picking for the receiving dock?,picking_optimization
|
| 304 |
What is a mini-load system?,general_faq
|
| 305 |
"A spill was reported near Conveyor-14 in the receiving dock, needs cleanup",safety_incident
|
| 306 |
Report vibration issue on Sorter-02,equipment_maintenance
|
| 307 |
+
Do we have enough SKU-1328 to fulfill 200 units?,inventory_check
|
| 308 |
Why do warehouses use cycle counting instead of annual counts?,general_faq
|
| 309 |
Route Crane-05 to picking station 12,agv_navigation
|
| 310 |
+
What's the fastest picking route for order #27681?,picking_optimization
|
| 311 |
Suggest a wave picking plan for the mezzanine,picking_optimization
|
| 312 |
Is the WCS connection to AGV-07 stable?,system_status
|
| 313 |
+
Do we have enough SKU-8264 to fulfill 200 units?,inventory_check
|
| 314 |
How should we sequence orders to minimize walking distance in the cross-dock area?,picking_optimization
|
| 315 |
Is the sorter in the cross-dock area running normally?,system_status
|
| 316 |
+
Track order #47175 for me,order_status
|
| 317 |
Pause Sorter-06 until the aisle in Zone E is clear,agv_navigation
|
| 318 |
Check system health for Zone B,system_status
|
| 319 |
Pause Crane-03 until the aisle in the cross-dock area is clear,agv_navigation
|
| 320 |
The sorter in Zone F keeps jamming,equipment_maintenance
|
| 321 |
+
What's the status of order #45116?,order_status
|
| 322 |
Optimize the pick path for Zone F,picking_optimization
|
| 323 |
Send AGV AMR-21 to Zone D,agv_navigation
|
| 324 |
+
Show the fulfillment status of #10654,order_status
|
| 325 |
+
Check inventory count for SKU-5294 in Zone C,inventory_check
|
| 326 |
+
What is the current stock level for SKU-9342?,inventory_check
|
| 327 |
What's the most efficient picking sequence for the cross-dock area today?,picking_optimization
|
| 328 |
"AMR-33 seems to be idle near Zone E, please reroute it",agv_navigation
|
| 329 |
Has AMR-33 reported any faults today?,system_status
|
| 330 |
+
Give me stock levels across all zones for SKU-8783,inventory_check
|
| 331 |
What is the uptime for AMR-33 today?,system_status
|
| 332 |
Route Crane-03 to picking station 18,agv_navigation
|
| 333 |
How can we reduce travel time for pickers in Zone F?,picking_optimization
|
| 334 |
+
What's the fastest picking route for order #65609?,picking_optimization
|
| 335 |
Report vibration issue on AMR-33,equipment_maintenance
|
| 336 |
Check system health for Zone E,system_status
|
| 337 |
"Belt AGV-07 stopped unexpectedly, please check",equipment_maintenance
|
| 338 |
+
Is order #76348 delayed?,order_status
|
| 339 |
The sorter in the shipping dock keeps jamming,equipment_maintenance
|
| 340 |
Report vibration issue on Conveyor-21,equipment_maintenance
|
| 341 |
+
Give me stock levels across all zones for SKU-9446,inventory_check
|
| 342 |
"A spill was reported near AMR-21 in the cross-dock area, needs cleanup",safety_incident
|
| 343 |
+
Show the fulfillment status of #76552,order_status
|
| 344 |
Please log a near-miss between a pedestrian and Conveyor-21 in the receiving dock,safety_incident
|
| 345 |
+
Is SKU-6228 in stock at the cross-dock area?,inventory_check
|
| 346 |
"Belt Conveyor-14 stopped unexpectedly, please check",equipment_maintenance
|
| 347 |
+
What is the current stock level for SKU-1173?,inventory_check
|
| 348 |
Are all cranes online in Zone B?,system_status
|
| 349 |
"A worker slipped near AMR-21, please log it",safety_incident
|
| 350 |
What is the current location of AGV-18?,agv_navigation
|
|
|
|
| 364 |
Would zone picking work better than batch picking for Zone E?,picking_optimization
|
| 365 |
Send the next available AGV to picking station 3,agv_navigation
|
| 366 |
Crane-03 motor temperature seems high,equipment_maintenance
|
| 367 |
+
Is SKU-6758 in stock at the receiving dock?,inventory_check
|
| 368 |
Why is AGV-12 stuck near the shipping dock?,agv_navigation
|
| 369 |
Send the next available AGV to picking station 5,agv_navigation
|
| 370 |
Please log a near-miss between a pedestrian and AGV-18 in Zone C,safety_incident
|
| 371 |
What's the most efficient picking sequence for Zone C today?,picking_optimization
|
| 372 |
+
Track order #51679 for me,order_status
|
| 373 |
+
Is SKU-1282 in stock at Zone A?,inventory_check
|
| 374 |
+
Do we have enough SKU-8718 to fulfill 200 units?,inventory_check
|
| 375 |
How does regenerative braking work on an AS/RS crane?,general_faq
|
| 376 |
+
Give me stock levels across all zones for SKU-9735,inventory_check
|
| 377 |
+
When will order #80380 be delivered?,order_status
|
| 378 |
Why is AGV-12 stuck near the receiving dock?,agv_navigation
|
| 379 |
+
Give me stock levels across all zones for SKU-5603,inventory_check
|
| 380 |
Optimize the pick path for Zone D,picking_optimization
|
| 381 |
Give me the current status of the WMS integration,system_status
|
| 382 |
+
When will order #47645 be delivered?,order_status
|
| 383 |
A forklift near-miss was reported in Zone F,safety_incident
|
| 384 |
+
What is the current stock level for SKU-1651?,inventory_check
|
| 385 |
"Someone bypassed the safety gate near AMR-21, please log this",safety_incident
|
| 386 |
How can we reduce travel time for pickers in Zone D?,picking_optimization
|
| 387 |
How can we reduce travel time for pickers in the shipping dock?,picking_optimization
|
| 388 |
+
Give me stock levels across all zones for SKU-9634,inventory_check
|
| 389 |
"A spill was reported near AGV-12 in Zone F, needs cleanup",safety_incident
|
| 390 |
+
Track order #91823 for me,order_status
|
| 391 |
Are all cranes online in Zone E?,system_status
|
| 392 |
Would zone picking work better than batch picking for Zone B?,picking_optimization
|
| 393 |
The sorter in the mezzanine keeps jamming,equipment_maintenance
|
|
|
|
| 397 |
A forklift near-miss was reported in Zone A,safety_incident
|
| 398 |
What is predictive maintenance?,general_faq
|
| 399 |
There was a near collision between Conveyor-14 and a pedestrian in Zone B,safety_incident
|
| 400 |
+
What's the fastest picking route for order #13867?,picking_optimization
|
| 401 |
+
Do we have enough SKU-8902 to fulfill 200 units?,inventory_check
|
| 402 |
Are all cranes online in the mezzanine?,system_status
|
| 403 |
+
Do we have enough SKU-7057 to fulfill 200 units?,inventory_check
|
| 404 |
Crane AMR-33 reported a fault code,equipment_maintenance
|
| 405 |
Pause Conveyor-14 until the aisle in Zone E is clear,agv_navigation
|
| 406 |
What's the difference between a stacker crane and a shuttle?,general_faq
|
| 407 |
Redirect Sorter-06 around the blocked aisle in Zone C,agv_navigation
|
| 408 |
Conveyor-21 motor temperature seems high,equipment_maintenance
|
| 409 |
+
Do we have enough SKU-5143 to fulfill 200 units?,inventory_check
|
| 410 |
+
Check inventory count for SKU-8395 in Zone E,inventory_check
|
| 411 |
Report unsafe pallet stacking in Zone C,safety_incident
|
| 412 |
Redirect Crane-09 around the blocked aisle in the mezzanine,agv_navigation
|
| 413 |
How should we sequence orders to minimize walking distance in the mezzanine?,picking_optimization
|
| 414 |
+
Do we have enough SKU-1218 to fulfill 200 units?,inventory_check
|
| 415 |
Please log a near-miss between a pedestrian and Crane-05 in the mezzanine,safety_incident
|
| 416 |
The conveyor belt in Zone F is making noise,equipment_maintenance
|
| 417 |
+
What's the fastest picking route for order #30359?,picking_optimization
|
| 418 |
Can you dispatch Conveyor-14 to the receiving dock right away?,agv_navigation
|
| 419 |
What's the current uptime percentage for the receiving dock?,system_status
|
| 420 |
Optimize the pick path for Zone B,picking_optimization
|
|
|
|
| 426 |
The conveyor belt in the shipping dock is making noise,equipment_maintenance
|
| 427 |
Log a breakdown for Crane-05 in the receiving dock,equipment_maintenance
|
| 428 |
There was a near collision between Sorter-06 and a pedestrian in Zone E,safety_incident
|
| 429 |
+
What's the status of order #26919?,order_status
|
| 430 |
What's the current uptime percentage for Zone B?,system_status
|
| 431 |
+
Track order #24084 for me,order_status
|
| 432 |
"A spill was reported near AGV-07 in the receiving dock, needs cleanup",safety_incident
|
| 433 |
Pause AGV-07 until the aisle in the cross-dock area is clear,agv_navigation
|
| 434 |
AGV-07 motor temperature seems high,equipment_maintenance
|
| 435 |
What KPIs matter most in warehouse automation?,general_faq
|
| 436 |
+
What is the current stock level for SKU-4486?,inventory_check
|
| 437 |
Redirect Crane-09 around the blocked aisle in Zone E,agv_navigation
|
| 438 |
+
Is order #97953 delayed?,order_status
|
| 439 |
+
Has order #64749 shipped yet?,order_status
|
| 440 |
+
What's the fastest picking route for order #21365?,picking_optimization
|
| 441 |
There was a near collision between AGV-12 and a pedestrian in Zone C,safety_incident
|
| 442 |
What is a pick rate and how is it measured?,general_faq
|
| 443 |
+
Why hasn't order #59121 left the dock yet?,order_status
|
data/latency_eval.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
-
"intent_classifier_ms": 0.
|
| 3 |
-
"anomaly_detector_ms":
|
| 4 |
-
"kb_retrieval_ms": 0.
|
| 5 |
"note": "LLM generation latency depends on the external Inference API call and is measured live in the app, not benchmarked here."
|
| 6 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"intent_classifier_ms": 0.476,
|
| 3 |
+
"anomaly_detector_ms": 10.104,
|
| 4 |
+
"kb_retrieval_ms": 0.735,
|
| 5 |
"note": "LLM generation latency depends on the external Inference API call and is measured live in the app, not benchmarked here."
|
| 6 |
}
|
data/retrieval_eval.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
{
|
| 2 |
-
"hit_rate_at_1": 0.
|
| 3 |
"hit_rate_at_2": 1.0,
|
| 4 |
-
"n_queries":
|
| 5 |
"rows": [
|
| 6 |
{
|
| 7 |
"query": "How does an AS/RS crane retrieve a pallet?",
|
|
@@ -9,7 +9,7 @@
|
|
| 9 |
"retrieved_top1": "asrs_overview",
|
| 10 |
"hit@1": true,
|
| 11 |
"hit@2": true,
|
| 12 |
-
"top1_score": 0.
|
| 13 |
},
|
| 14 |
{
|
| 15 |
"query": "What's the difference between an AGV and an AMR?",
|
|
@@ -17,7 +17,7 @@
|
|
| 17 |
"retrieved_top1": "safety_protocol",
|
| 18 |
"hit@1": false,
|
| 19 |
"hit@2": true,
|
| 20 |
-
"top1_score": 0.
|
| 21 |
},
|
| 22 |
{
|
| 23 |
"query": "What does a WMS integrate with?",
|
|
@@ -25,7 +25,7 @@
|
|
| 25 |
"retrieved_top1": "wms_overview",
|
| 26 |
"hit@1": true,
|
| 27 |
"hit@2": true,
|
| 28 |
-
"top1_score": 0.
|
| 29 |
},
|
| 30 |
{
|
| 31 |
"query": "Why would a sorter jam?",
|
|
@@ -33,7 +33,7 @@
|
|
| 33 |
"retrieved_top1": "conveyor_sorting",
|
| 34 |
"hit@1": true,
|
| 35 |
"hit@2": true,
|
| 36 |
-
"top1_score": 0.
|
| 37 |
},
|
| 38 |
{
|
| 39 |
"query": "What is batch picking?",
|
|
@@ -41,7 +41,7 @@
|
|
| 41 |
"retrieved_top1": "picking_strategies",
|
| 42 |
"hit@1": true,
|
| 43 |
"hit@2": true,
|
| 44 |
-
"top1_score": 0.
|
| 45 |
},
|
| 46 |
{
|
| 47 |
"query": "How can we predict a motor failure before it happens?",
|
|
@@ -49,7 +49,7 @@
|
|
| 49 |
"retrieved_top1": "conveyor_sorting",
|
| 50 |
"hit@1": false,
|
| 51 |
"hit@2": true,
|
| 52 |
-
"top1_score": 0.
|
| 53 |
},
|
| 54 |
{
|
| 55 |
"query": "What should I do after a near-miss with a forklift?",
|
|
@@ -57,7 +57,7 @@
|
|
| 57 |
"retrieved_top1": "safety_protocol",
|
| 58 |
"hit@1": true,
|
| 59 |
"hit@2": true,
|
| 60 |
-
"top1_score": 0.
|
| 61 |
},
|
| 62 |
{
|
| 63 |
"query": "How do we keep inventory counts accurate?",
|
|
@@ -65,7 +65,7 @@
|
|
| 65 |
"retrieved_top1": "inventory_accuracy",
|
| 66 |
"hit@1": true,
|
| 67 |
"hit@2": true,
|
| 68 |
-
"top1_score": 0.
|
| 69 |
},
|
| 70 |
{
|
| 71 |
"query": "What KPIs should a warehouse manager track?",
|
|
@@ -73,7 +73,7 @@
|
|
| 73 |
"retrieved_top1": "kpi_overview",
|
| 74 |
"hit@1": true,
|
| 75 |
"hit@2": true,
|
| 76 |
-
"top1_score": 0.
|
| 77 |
},
|
| 78 |
{
|
| 79 |
"query": "How can automated warehouses save energy?",
|
|
@@ -81,7 +81,63 @@
|
|
| 81 |
"retrieved_top1": "energy_efficiency",
|
| 82 |
"hit@1": true,
|
| 83 |
"hit@2": true,
|
| 84 |
-
"top1_score": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 85 |
}
|
| 86 |
]
|
| 87 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"hit_rate_at_1": 0.8823529411764706,
|
| 3 |
"hit_rate_at_2": 1.0,
|
| 4 |
+
"n_queries": 17,
|
| 5 |
"rows": [
|
| 6 |
{
|
| 7 |
"query": "How does an AS/RS crane retrieve a pallet?",
|
|
|
|
| 9 |
"retrieved_top1": "asrs_overview",
|
| 10 |
"hit@1": true,
|
| 11 |
"hit@2": true,
|
| 12 |
+
"top1_score": 0.2129
|
| 13 |
},
|
| 14 |
{
|
| 15 |
"query": "What's the difference between an AGV and an AMR?",
|
|
|
|
| 17 |
"retrieved_top1": "safety_protocol",
|
| 18 |
"hit@1": false,
|
| 19 |
"hit@2": true,
|
| 20 |
+
"top1_score": 0.1543
|
| 21 |
},
|
| 22 |
{
|
| 23 |
"query": "What does a WMS integrate with?",
|
|
|
|
| 25 |
"retrieved_top1": "wms_overview",
|
| 26 |
"hit@1": true,
|
| 27 |
"hit@2": true,
|
| 28 |
+
"top1_score": 0.2609
|
| 29 |
},
|
| 30 |
{
|
| 31 |
"query": "Why would a sorter jam?",
|
|
|
|
| 33 |
"retrieved_top1": "conveyor_sorting",
|
| 34 |
"hit@1": true,
|
| 35 |
"hit@2": true,
|
| 36 |
+
"top1_score": 0.0958
|
| 37 |
},
|
| 38 |
{
|
| 39 |
"query": "What is batch picking?",
|
|
|
|
| 41 |
"retrieved_top1": "picking_strategies",
|
| 42 |
"hit@1": true,
|
| 43 |
"hit@2": true,
|
| 44 |
+
"top1_score": 0.3191
|
| 45 |
},
|
| 46 |
{
|
| 47 |
"query": "How can we predict a motor failure before it happens?",
|
|
|
|
| 49 |
"retrieved_top1": "conveyor_sorting",
|
| 50 |
"hit@1": false,
|
| 51 |
"hit@2": true,
|
| 52 |
+
"top1_score": 0.1501
|
| 53 |
},
|
| 54 |
{
|
| 55 |
"query": "What should I do after a near-miss with a forklift?",
|
|
|
|
| 57 |
"retrieved_top1": "safety_protocol",
|
| 58 |
"hit@1": true,
|
| 59 |
"hit@2": true,
|
| 60 |
+
"top1_score": 0.3327
|
| 61 |
},
|
| 62 |
{
|
| 63 |
"query": "How do we keep inventory counts accurate?",
|
|
|
|
| 65 |
"retrieved_top1": "inventory_accuracy",
|
| 66 |
"hit@1": true,
|
| 67 |
"hit@2": true,
|
| 68 |
+
"top1_score": 0.1017
|
| 69 |
},
|
| 70 |
{
|
| 71 |
"query": "What KPIs should a warehouse manager track?",
|
|
|
|
| 73 |
"retrieved_top1": "kpi_overview",
|
| 74 |
"hit@1": true,
|
| 75 |
"hit@2": true,
|
| 76 |
+
"top1_score": 0.2761
|
| 77 |
},
|
| 78 |
{
|
| 79 |
"query": "How can automated warehouses save energy?",
|
|
|
|
| 81 |
"retrieved_top1": "energy_efficiency",
|
| 82 |
"hit@1": true,
|
| 83 |
"hit@2": true,
|
| 84 |
+
"top1_score": 0.4415
|
| 85 |
+
},
|
| 86 |
+
{
|
| 87 |
+
"query": "What is cross-docking and why use it?",
|
| 88 |
+
"expected": "cross_docking",
|
| 89 |
+
"retrieved_top1": "cross_docking",
|
| 90 |
+
"hit@1": true,
|
| 91 |
+
"hit@2": true,
|
| 92 |
+
"top1_score": 0.3365
|
| 93 |
+
},
|
| 94 |
+
{
|
| 95 |
+
"query": "How should a warehouse handle customer returns?",
|
| 96 |
+
"expected": "reverse_logistics",
|
| 97 |
+
"retrieved_top1": "reverse_logistics",
|
| 98 |
+
"hit@1": true,
|
| 99 |
+
"hit@2": true,
|
| 100 |
+
"top1_score": 0.2315
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"query": "Should I use RFID or barcodes for pallet tracking?",
|
| 104 |
+
"expected": "rfid_vs_barcode",
|
| 105 |
+
"retrieved_top1": "rfid_vs_barcode",
|
| 106 |
+
"hit@1": true,
|
| 107 |
+
"hit@2": true,
|
| 108 |
+
"top1_score": 0.2714
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"query": "How do I avoid congestion at the loading dock?",
|
| 112 |
+
"expected": "dock_scheduling",
|
| 113 |
+
"retrieved_top1": "dock_scheduling",
|
| 114 |
+
"hit@1": true,
|
| 115 |
+
"hit@2": true,
|
| 116 |
+
"top1_score": 0.2939
|
| 117 |
+
},
|
| 118 |
+
{
|
| 119 |
+
"query": "How should I decide where to store a new SKU?",
|
| 120 |
+
"expected": "slotting_optimization",
|
| 121 |
+
"retrieved_top1": "slotting_optimization",
|
| 122 |
+
"hit@1": true,
|
| 123 |
+
"hit@2": true,
|
| 124 |
+
"top1_score": 0.1115
|
| 125 |
+
},
|
| 126 |
+
{
|
| 127 |
+
"query": "How do warehouses control access to high-value inventory?",
|
| 128 |
+
"expected": "warehouse_security",
|
| 129 |
+
"retrieved_top1": "warehouse_security",
|
| 130 |
+
"hit@1": true,
|
| 131 |
+
"hit@2": true,
|
| 132 |
+
"top1_score": 0.4407
|
| 133 |
+
},
|
| 134 |
+
{
|
| 135 |
+
"query": "How can I tell if a picker is underperforming or just has a hard task?",
|
| 136 |
+
"expected": "labor_management",
|
| 137 |
+
"retrieved_top1": "labor_management",
|
| 138 |
+
"hit@1": true,
|
| 139 |
+
"hit@2": true,
|
| 140 |
+
"top1_score": 0.081
|
| 141 |
}
|
| 142 |
]
|
| 143 |
}
|
data/sample_sensor_batch.csv
ADDED
|
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
id,motor_temp_c,vibration_mm_s,current_amps,belt_speed_mps
|
| 2 |
+
R001,53.17,1.6,10.33,1.53
|
| 3 |
+
R002,72.15,3.97,18.18,0.66
|
| 4 |
+
R003,77.61,5.44,21.41,0.79
|
| 5 |
+
R004,47.4,2.24,11.98,1.51
|
| 6 |
+
R005,51.09,2.65,11.54,1.69
|
| 7 |
+
R006,56.19,1.8,10.81,1.84
|
| 8 |
+
R007,49.93,2.17,11.96,1.33
|
| 9 |
+
R008,53.9,2.18,12.97,1.38
|
| 10 |
+
R009,52.52,2.24,11.45,1.58
|
| 11 |
+
R010,77.04,4.75,17.26,0.49
|
| 12 |
+
R011,55.45,1.61,12.52,1.62
|
| 13 |
+
R012,54.06,2.54,12.98,1.2
|
| 14 |
+
R013,51.28,2.26,11.54,1.34
|
| 15 |
+
R014,52.85,1.97,12.88,1.61
|
| 16 |
+
R015,51.03,2.14,11.85,1.57
|
| 17 |
+
R016,57.78,3.2,11.82,1.37
|
| 18 |
+
R017,47.63,2.11,13.75,1.48
|
| 19 |
+
R018,55.63,2.92,11.49,1.18
|
| 20 |
+
R019,56.43,1.43,14.29,1.49
|
| 21 |
+
R020,55.0,2.73,10.01,1.52
|
| 22 |
+
R021,55.42,2.63,11.36,1.51
|
| 23 |
+
R022,48.88,1.91,9.47,1.52
|
| 24 |
+
R023,49.62,2.58,11.7,1.65
|
| 25 |
+
R024,74.74,5.32,19.02,0.7
|
| 26 |
+
R025,55.24,2.26,13.89,1.47
|
| 27 |
+
R026,60.36,2.23,13.03,1.47
|
| 28 |
+
R027,53.09,2.1,8.95,1.27
|
| 29 |
+
R028,44.93,2.3,11.99,1.24
|
| 30 |
+
R029,56.96,2.88,11.62,1.34
|
| 31 |
+
R030,51.76,2.77,10.65,1.72
|
| 32 |
+
R031,84.59,7.01,17.6,1.4
|
| 33 |
+
R032,53.03,1.59,11.51,1.61
|
| 34 |
+
R033,76.36,4.55,19.03,0.81
|
| 35 |
+
R034,54.88,1.88,12.53,1.54
|
| 36 |
+
R035,54.81,2.26,10.06,1.37
|
| 37 |
+
R036,56.09,2.53,13.0,1.55
|
| 38 |
+
R037,49.84,2.49,11.33,1.41
|
| 39 |
+
R038,51.44,2.64,9.01,1.41
|
| 40 |
+
R039,54.25,1.86,13.58,1.63
|
| 41 |
+
R040,53.18,1.91,11.31,1.53
|
data/vision_eval.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"standard_noise_level": 0.4,
|
| 3 |
+
"n_test": 100,
|
| 4 |
+
"exact_match_rate": 0.88,
|
| 5 |
+
"count_mae": 0.17,
|
| 6 |
+
"occupancy_mae": 0.009444444444444445,
|
| 7 |
+
"noise_sweep_levels": [
|
| 8 |
+
0.0,
|
| 9 |
+
0.2,
|
| 10 |
+
0.4,
|
| 11 |
+
0.6,
|
| 12 |
+
0.8,
|
| 13 |
+
1.0
|
| 14 |
+
],
|
| 15 |
+
"noise_sweep_exact_match_rates": [
|
| 16 |
+
1.0,
|
| 17 |
+
0.9166666666666666,
|
| 18 |
+
0.8833333333333333,
|
| 19 |
+
0.7666666666666667,
|
| 20 |
+
0.8333333333333334,
|
| 21 |
+
0.7666666666666667
|
| 22 |
+
],
|
| 23 |
+
"method": "Classical CV: color-distance segmentation -> morphology -> connected components -> geometric filtering. No pretrained weights."
|
| 24 |
+
}
|
models/anomaly_iforest.joblib
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 2166973
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0e238af00762e3dced90d74fe626523c83cf0cf3ecd2b7a8c6931ace04ab1641
|
| 3 |
size 2166973
|
models/anomaly_scaler.joblib
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 679
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:68bbb35009b07935b7cfd67fc41d04f8189ab81768d5ec57859c1ca139bff73e
|
| 3 |
size 679
|
models/intent_pipeline.joblib
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c6f690a27b696b3e825190eec1cbe9491def156ce40347146fc8cd8714799959
|
| 3 |
+
size 86780
|
requirements.txt
CHANGED
|
@@ -1,8 +1,11 @@
|
|
| 1 |
gradio==5.9.1
|
|
|
|
| 2 |
scikit-learn==1.7.2
|
| 3 |
pandas==2.2.3
|
| 4 |
numpy==1.26.4
|
| 5 |
matplotlib==3.9.2
|
| 6 |
joblib==1.4.2
|
| 7 |
huggingface_hub==0.31.4
|
|
|
|
|
|
|
| 8 |
spaces>=0.30.0
|
|
|
|
| 1 |
gradio==5.9.1
|
| 2 |
+
pydantic==2.10.6
|
| 3 |
scikit-learn==1.7.2
|
| 4 |
pandas==2.2.3
|
| 5 |
numpy==1.26.4
|
| 6 |
matplotlib==3.9.2
|
| 7 |
joblib==1.4.2
|
| 8 |
huggingface_hub==0.31.4
|
| 9 |
+
opencv-python-headless==4.10.0.84
|
| 10 |
+
requests>=2.31.0
|
| 11 |
spaces>=0.30.0
|
src/alerting.py
ADDED
|
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
alerting.py -- ops integration: Slack alert (real Incoming Webhook if the
|
| 3 |
+
`SLACK_WEBHOOK_URL` secret is set) with a clearly-labelled preview fallback
|
| 4 |
+
otherwise, plus an email preview. Mirrors llm_client.py's pattern: real
|
| 5 |
+
integration when configured, never silently pretends to have sent anything.
|
| 6 |
+
"""
|
| 7 |
+
import os
|
| 8 |
+
from dataclasses import dataclass
|
| 9 |
+
from datetime import datetime, timezone
|
| 10 |
+
from typing import Optional
|
| 11 |
+
|
| 12 |
+
try:
|
| 13 |
+
import requests
|
| 14 |
+
except ImportError: # pragma: no cover
|
| 15 |
+
requests = None
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
@dataclass
|
| 19 |
+
class AlertResult:
|
| 20 |
+
sent: bool
|
| 21 |
+
channel: str # "slack" | "preview-only"
|
| 22 |
+
message_preview: str
|
| 23 |
+
error: Optional[str] = None
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
def _fmt(zone: str, r: dict, score: float) -> str:
|
| 27 |
+
ts = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M UTC")
|
| 28 |
+
return (
|
| 29 |
+
f":rotating_light: *Equipment anomaly detected* -- {zone}\n"
|
| 30 |
+
f"*Time:* {ts}\n*Anomaly score:* {score:.2f} / 1.00\n"
|
| 31 |
+
f"*Readings:* temp={r.get('motor_temp_c','?')}°C, vibration={r.get('vibration_mm_s','?')} mm/s, "
|
| 32 |
+
f"current={r.get('current_amps','?')} A, belt_speed={r.get('belt_speed_mps','?')} m/s\n"
|
| 33 |
+
f"*Recommended action:* flag for inspection before next shift (bearing wear / "
|
| 34 |
+
f"misalignment / motor overload pattern)."
|
| 35 |
+
)
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
def send_anomaly_alert(zone: str, reading: dict, score: float) -> AlertResult:
|
| 39 |
+
text = _fmt(zone, reading, score)
|
| 40 |
+
url = os.environ.get("SLACK_WEBHOOK_URL")
|
| 41 |
+
if not url:
|
| 42 |
+
return AlertResult(False, "preview-only", text)
|
| 43 |
+
if requests is None:
|
| 44 |
+
return AlertResult(False, "preview-only", text, "`requests` not installed")
|
| 45 |
+
try:
|
| 46 |
+
resp = requests.post(url, json={"text": text}, timeout=5)
|
| 47 |
+
if resp.status_code == 200:
|
| 48 |
+
return AlertResult(True, "slack", text)
|
| 49 |
+
return AlertResult(False, "preview-only", text, f"Slack HTTP {resp.status_code}: {resp.text[:200]}")
|
| 50 |
+
except Exception as e: # noqa: BLE001
|
| 51 |
+
return AlertResult(False, "preview-only", text, f"{type(e).__name__}: {e}")
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
def format_email_preview(zone: str, reading: dict, score: float) -> str:
|
| 55 |
+
ts = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M UTC")
|
| 56 |
+
return (
|
| 57 |
+
f"To: maintenance-team@example.com\nSubject: [ALERT] Equipment anomaly detected -- {zone}\n\n"
|
| 58 |
+
f"Detected at {ts}. Anomaly score: {score:.2f} / 1.00\n"
|
| 59 |
+
f"Temp {reading.get('motor_temp_c','?')} C | Vibration {reading.get('vibration_mm_s','?')} mm/s | "
|
| 60 |
+
f"Current {reading.get('current_amps','?')} A | Belt {reading.get('belt_speed_mps','?')} m/s\n"
|
| 61 |
+
f"Recommended action: flag for inspection before the next shift.\n"
|
| 62 |
+
)
|
src/data_generation.py
CHANGED
|
@@ -268,4 +268,11 @@ RETRIEVAL_EVAL_SET = [
|
|
| 268 |
("How do we keep inventory counts accurate?", "inventory_accuracy"),
|
| 269 |
("What KPIs should a warehouse manager track?", "kpi_overview"),
|
| 270 |
("How can automated warehouses save energy?", "energy_efficiency"),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 271 |
]
|
|
|
|
| 268 |
("How do we keep inventory counts accurate?", "inventory_accuracy"),
|
| 269 |
("What KPIs should a warehouse manager track?", "kpi_overview"),
|
| 270 |
("How can automated warehouses save energy?", "energy_efficiency"),
|
| 271 |
+
("What is cross-docking and why use it?", "cross_docking"),
|
| 272 |
+
("How should a warehouse handle customer returns?", "reverse_logistics"),
|
| 273 |
+
("Should I use RFID or barcodes for pallet tracking?", "rfid_vs_barcode"),
|
| 274 |
+
("How do I avoid congestion at the loading dock?", "dock_scheduling"),
|
| 275 |
+
("How should I decide where to store a new SKU?", "slotting_optimization"),
|
| 276 |
+
("How do warehouses control access to high-value inventory?", "warehouse_security"),
|
| 277 |
+
("How can I tell if a picker is underperforming or just has a hard task?", "labor_management"),
|
| 278 |
]
|
src/knowledge_base.py
CHANGED
|
@@ -155,4 +155,59 @@ KNOWLEDGE_BASE = [
|
|
| 155 |
"electricity tariff windows."
|
| 156 |
),
|
| 157 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 158 |
]
|
|
|
|
| 155 |
"electricity tariff windows."
|
| 156 |
),
|
| 157 |
},
|
| 158 |
+
{
|
| 159 |
+
"id": "cross_docking", "title": "Cross-Docking", "category": "process",
|
| 160 |
+
"text": ("Cross-docking moves inbound goods directly from receiving to an outbound "
|
| 161 |
+
"shipment with little or no intermediate storage, typically staged for only a "
|
| 162 |
+
"few hours. It reduces inventory holding cost and handling time but requires "
|
| 163 |
+
"tightly synchronized inbound and outbound schedules, since a delayed inbound "
|
| 164 |
+
"truck can cascade into missed outbound departures."),
|
| 165 |
+
},
|
| 166 |
+
{
|
| 167 |
+
"id": "reverse_logistics", "title": "Returns & Reverse Logistics", "category": "process",
|
| 168 |
+
"text": ("Reverse logistics covers the flow of returned goods back through the warehouse: "
|
| 169 |
+
"inspection, grading (resellable, refurbishable, or scrap), restocking, and "
|
| 170 |
+
"sometimes vendor chargebacks. High e-commerce return rates make a dedicated "
|
| 171 |
+
"returns processing area and fast grading turnaround important for recovering "
|
| 172 |
+
"inventory value quickly."),
|
| 173 |
+
},
|
| 174 |
+
{
|
| 175 |
+
"id": "rfid_vs_barcode", "title": "RFID vs. Barcode Scanning", "category": "equipment",
|
| 176 |
+
"text": ("Barcode scanning requires line-of-sight to a single label and is cheap and highly "
|
| 177 |
+
"accurate. RFID (radio-frequency identification) can read many tags at once "
|
| 178 |
+
"without line-of-sight, speeding up cycle counting and dock receiving, but tags "
|
| 179 |
+
"cost more and dense metal racking can cause read interference. Many warehouses "
|
| 180 |
+
"use barcodes for unit-level picking and RFID for pallet- or case-level tracking."),
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"id": "dock_scheduling", "title": "Dock Scheduling & Yard Management", "category": "process",
|
| 184 |
+
"text": ("Dock scheduling assigns inbound and outbound trucks to specific loading dock doors "
|
| 185 |
+
"and time slots to avoid yard congestion and dock-door bottlenecks. A yard "
|
| 186 |
+
"management system (YMS) tracks trailer location and status across the yard and "
|
| 187 |
+
"is often integrated with the WMS so put-away labor can be planned before a "
|
| 188 |
+
"trailer even arrives."),
|
| 189 |
+
},
|
| 190 |
+
{
|
| 191 |
+
"id": "slotting_optimization", "title": "Warehouse Slotting Optimization", "category": "process",
|
| 192 |
+
"text": ("Slotting assigns each SKU to a storage location based on pick frequency, size, "
|
| 193 |
+
"weight, and affinity with frequently co-ordered SKUs. Placing high-velocity items "
|
| 194 |
+
"near pack-out stations and keeping heavy items at waist height are common rules. "
|
| 195 |
+
"Poor slotting is one of the most common causes of excess picker travel time."),
|
| 196 |
+
},
|
| 197 |
+
{
|
| 198 |
+
"id": "warehouse_security", "title": "Warehouse Security & Access Control", "category": "safety",
|
| 199 |
+
"text": ("Warehouse security typically combines badge-based access control at building and "
|
| 200 |
+
"high-value-cage entry points, CCTV coverage of dock doors and aisles, and "
|
| 201 |
+
"shrinkage tracking (the gap between recorded and physically counted inventory). "
|
| 202 |
+
"High-value or regulated goods often require a separate caged area with its own "
|
| 203 |
+
"access log."),
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"id": "labor_management", "title": "Labor Management & Workforce Scheduling", "category": "process",
|
| 207 |
+
"text": ("A labor management system (LMS) sets expected task times based on engineered "
|
| 208 |
+
"standards, tracks actual performance against them, and helps schedule headcount "
|
| 209 |
+
"to forecast order volume by shift. It flags tasks taking significantly longer "
|
| 210 |
+
"than expected, which can indicate a process problem or a training gap rather "
|
| 211 |
+
"than a motivation problem."),
|
| 212 |
+
},
|
| 213 |
]
|
src/llm_client.py
CHANGED
|
@@ -107,15 +107,40 @@ def _extractive_fallback(query: str, sources: List[RetrievedDoc]) -> str:
|
|
| 107 |
)
|
| 108 |
|
| 109 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 110 |
def answer_query(
|
| 111 |
query: str,
|
| 112 |
retriever: KBRetriever,
|
| 113 |
k: int = 2,
|
| 114 |
max_tokens: int = 350,
|
|
|
|
| 115 |
) -> AssistantResponse:
|
|
|
|
|
|
|
| 116 |
start = time.time()
|
| 117 |
sources = retriever.retrieve(query, k=k)
|
| 118 |
context_block = "\n\n".join(f"[{s.title}]\n{s.text}" for s in sources)
|
|
|
|
| 119 |
|
| 120 |
hf_token = _get_hf_token()
|
| 121 |
|
|
@@ -144,10 +169,9 @@ def answer_query(
|
|
| 144 |
debug_errors=[f"huggingface_hub not importable: {e}"],
|
| 145 |
)
|
| 146 |
|
| 147 |
-
messages = [
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
]
|
| 151 |
|
| 152 |
errors = []
|
| 153 |
for candidate in MODEL_CANDIDATES:
|
|
|
|
| 107 |
)
|
| 108 |
|
| 109 |
|
| 110 |
+
MAX_HISTORY_TURNS = 4 # prior (user, assistant) pairs carried into the LLM call
|
| 111 |
+
|
| 112 |
+
|
| 113 |
+
def _clean_history(history: Optional[List[dict]]) -> List[dict]:
|
| 114 |
+
"""Normalize Gradio's messages-format history into a short, clean list safe
|
| 115 |
+
to feed back to the LLM: strips our own metadata footer from assistant
|
| 116 |
+
turns and keeps only the last MAX_HISTORY_TURNS pairs (bounded prompt)."""
|
| 117 |
+
if not history:
|
| 118 |
+
return []
|
| 119 |
+
cleaned = []
|
| 120 |
+
for turn in history:
|
| 121 |
+
role, content = turn.get("role"), turn.get("content", "")
|
| 122 |
+
if not isinstance(content, str):
|
| 123 |
+
continue
|
| 124 |
+
if role == "assistant" and "\n\n---\n" in content:
|
| 125 |
+
content = content.split("\n\n---\n")[0]
|
| 126 |
+
if role in ("user", "assistant") and content.strip():
|
| 127 |
+
cleaned.append({"role": role, "content": content.strip()})
|
| 128 |
+
return cleaned[-MAX_HISTORY_TURNS * 2:]
|
| 129 |
+
|
| 130 |
+
|
| 131 |
def answer_query(
|
| 132 |
query: str,
|
| 133 |
retriever: KBRetriever,
|
| 134 |
k: int = 2,
|
| 135 |
max_tokens: int = 350,
|
| 136 |
+
history: Optional[List[dict]] = None,
|
| 137 |
) -> AssistantResponse:
|
| 138 |
+
"""`history`: prior turns (Gradio messages format) enabling multi-turn
|
| 139 |
+
follow-ups like "what about Zone D instead?". Omit for single-turn."""
|
| 140 |
start = time.time()
|
| 141 |
sources = retriever.retrieve(query, k=k)
|
| 142 |
context_block = "\n\n".join(f"[{s.title}]\n{s.text}" for s in sources)
|
| 143 |
+
prior_turns = _clean_history(history)
|
| 144 |
|
| 145 |
hf_token = _get_hf_token()
|
| 146 |
|
|
|
|
| 169 |
debug_errors=[f"huggingface_hub not importable: {e}"],
|
| 170 |
)
|
| 171 |
|
| 172 |
+
messages = [{"role": "system", "content": SYSTEM_PROMPT}]
|
| 173 |
+
messages.extend(prior_turns)
|
| 174 |
+
messages.append({"role": "user", "content": f"CONTEXT:\n{context_block}\n\nQUESTION: {query}"})
|
|
|
|
| 175 |
|
| 176 |
errors = []
|
| 177 |
for candidate in MODEL_CANDIDATES:
|
src/vision_model.py
ADDED
|
@@ -0,0 +1,164 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
vision_model.py
|
| 3 |
+
-----------------
|
| 4 |
+
Visual Inventory & Rack Occupancy Scanner.
|
| 5 |
+
|
| 6 |
+
A deliberately different computer-vision approach from a YOLO-style deep
|
| 7 |
+
detector: a small, fully deterministic **classical image-processing
|
| 8 |
+
pipeline** (OpenCV: color-based background/shelf segmentation ->
|
| 9 |
+
morphological cleanup -> connected components -> geometric filtering) that
|
| 10 |
+
(a) counts box/pallet-like objects on a warehouse rack and (b) estimates
|
| 11 |
+
what fraction of the rack's storage slots are occupied.
|
| 12 |
+
|
| 13 |
+
Why classical CV, deliberately: zero model downloads (self-contained and
|
| 14 |
+
reproducible), runs instantly on a free CPU Space, and every detection is a
|
| 15 |
+
traceable contour rather than an opaque activation. A production system
|
| 16 |
+
would likely fine-tune a detector; this module demonstrates the systems
|
| 17 |
+
thinking (synthetic ground truth, honest evaluation under noise) that would
|
| 18 |
+
carry over directly.
|
| 19 |
+
|
| 20 |
+
Synthetic scenes: rack photos can't be licensed/scraped for a portfolio
|
| 21 |
+
project, so -- like every other dataset here -- scenes are generated with
|
| 22 |
+
PIL with known ground truth, which is what allows honest metrics.
|
| 23 |
+
|
| 24 |
+
Development notes (kept because they're real):
|
| 25 |
+
* v1 used Canny edges + heavy dilation and merged every box and shelf bar
|
| 26 |
+
into one blob. Switching to per-pixel color-distance segmentation fixed it.
|
| 27 |
+
* On clean renders (noise_level=0) v2 hit a suspiciously perfect 100%, so
|
| 28 |
+
`noise_level` (Gaussian pixel noise, brightness/contrast jitter, per-box
|
| 29 |
+
color variation) was added; accuracy then degrades realistically.
|
| 30 |
+
* Occupancy was first computed from padded box pixel area vs full cell area
|
| 31 |
+
(a definitional mismatch that under-counted ~40%); it now counts occupied
|
| 32 |
+
grid cells, matching how ground truth is defined.
|
| 33 |
+
"""
|
| 34 |
+
|
| 35 |
+
import random
|
| 36 |
+
from dataclasses import dataclass
|
| 37 |
+
from typing import List, Optional, Tuple
|
| 38 |
+
|
| 39 |
+
import cv2
|
| 40 |
+
import numpy as np
|
| 41 |
+
from PIL import Image, ImageDraw
|
| 42 |
+
|
| 43 |
+
RACK_W, RACK_H = 640, 400
|
| 44 |
+
SHELF_ROWS, SHELF_COLS = 3, 6
|
| 45 |
+
MARGIN = 20
|
| 46 |
+
SHELF_BAR_H = 10
|
| 47 |
+
BOX_COLORS = [
|
| 48 |
+
(181, 101, 29), (139, 69, 19), (205, 133, 63), (160, 82, 45),
|
| 49 |
+
(222, 184, 135), (128, 0, 0), (46, 87, 46), (25, 65, 100),
|
| 50 |
+
]
|
| 51 |
+
BG_COLOR = (235, 235, 230)
|
| 52 |
+
SHELF_COLOR = (90, 90, 95)
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
@dataclass
|
| 56 |
+
class RackScene:
|
| 57 |
+
image: Image.Image
|
| 58 |
+
true_box_count: int
|
| 59 |
+
true_occupancy: float # fraction of cells filled, 0-1
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
def generate_rack_scene(fill_prob: float = 0.6, seed: Optional[int] = None, noise_level: float = 0.0) -> RackScene:
|
| 63 |
+
"""One synthetic rack image with known ground-truth box count/occupancy."""
|
| 64 |
+
rng = random.Random(seed)
|
| 65 |
+
img = Image.new("RGB", (RACK_W, RACK_H), BG_COLOR)
|
| 66 |
+
draw = ImageDraw.Draw(img)
|
| 67 |
+
cell_w = (RACK_W - 2 * MARGIN) / SHELF_COLS
|
| 68 |
+
cell_h = (RACK_H - 2 * MARGIN) / SHELF_ROWS
|
| 69 |
+
|
| 70 |
+
for row in range(SHELF_ROWS + 1):
|
| 71 |
+
y = MARGIN + row * cell_h
|
| 72 |
+
draw.rectangle([MARGIN - 5, y - SHELF_BAR_H / 2, RACK_W - MARGIN + 5, y + SHELF_BAR_H / 2], fill=SHELF_COLOR)
|
| 73 |
+
|
| 74 |
+
n_filled = 0
|
| 75 |
+
for row in range(SHELF_ROWS):
|
| 76 |
+
for col in range(SHELF_COLS):
|
| 77 |
+
if rng.random() < fill_prob:
|
| 78 |
+
x0, y0 = MARGIN + col * cell_w, MARGIN + row * cell_h
|
| 79 |
+
pad = min(cell_w, cell_h) * 0.12
|
| 80 |
+
bw, bh = cell_w - 2 * pad, cell_h - 2 * pad - 4
|
| 81 |
+
bx0, by0 = x0 + pad, y0 + pad + 4
|
| 82 |
+
base = rng.choice(BOX_COLORS)
|
| 83 |
+
if noise_level > 0:
|
| 84 |
+
j = int(25 * noise_level)
|
| 85 |
+
color = tuple(int(np.clip(c + rng.randint(-j, j), 0, 255)) for c in base)
|
| 86 |
+
else:
|
| 87 |
+
color = base
|
| 88 |
+
draw.rectangle([bx0, by0, bx0 + bw, by0 + bh], fill=color, outline=(30, 30, 30), width=2)
|
| 89 |
+
draw.line([bx0 + 4, by0 + bh * 0.4, bx0 + bw - 4, by0 + bh * 0.4], fill=(20, 20, 20), width=2)
|
| 90 |
+
n_filled += 1
|
| 91 |
+
|
| 92 |
+
if noise_level > 0:
|
| 93 |
+
arr = np.array(img).astype(np.float32)
|
| 94 |
+
brightness = rng.uniform(-20, 20) * noise_level
|
| 95 |
+
contrast = 1.0 + rng.uniform(-0.15, 0.15) * noise_level
|
| 96 |
+
arr = (arr - 127.5) * contrast + 127.5 + brightness
|
| 97 |
+
arr += np.random.RandomState(seed).normal(0, 10 * noise_level, arr.shape)
|
| 98 |
+
img = Image.fromarray(np.clip(arr, 0, 255).astype(np.uint8))
|
| 99 |
+
|
| 100 |
+
return RackScene(img, n_filled, n_filled / (SHELF_ROWS * SHELF_COLS))
|
| 101 |
+
|
| 102 |
+
|
| 103 |
+
@dataclass
|
| 104 |
+
class DetectionResult:
|
| 105 |
+
detected_count: int
|
| 106 |
+
estimated_occupancy: float
|
| 107 |
+
annotated_image: Image.Image
|
| 108 |
+
boxes: List[Tuple[int, int, int, int]]
|
| 109 |
+
|
| 110 |
+
|
| 111 |
+
def _iou(a, b) -> float:
|
| 112 |
+
ax, ay, aw, ah = a
|
| 113 |
+
bx, by, bw, bh = b
|
| 114 |
+
iw = max(0, min(ax + aw, bx + bw) - max(ax, bx))
|
| 115 |
+
ih = max(0, min(ay + ah, by + bh) - max(ay, by))
|
| 116 |
+
inter = iw * ih
|
| 117 |
+
union = aw * ah + bw * bh - inter
|
| 118 |
+
return inter / union if union > 0 else 0.0
|
| 119 |
+
|
| 120 |
+
|
| 121 |
+
def _nms(boxes, thr=0.3):
|
| 122 |
+
kept = []
|
| 123 |
+
for b in sorted(boxes, key=lambda b: b[2] * b[3], reverse=True):
|
| 124 |
+
if all(_iou(b, k) < thr for k in kept):
|
| 125 |
+
kept.append(b)
|
| 126 |
+
return kept
|
| 127 |
+
|
| 128 |
+
|
| 129 |
+
def detect_boxes(image: Image.Image, min_area_frac: float = 0.01, max_area_frac: float = 0.25) -> DetectionResult:
|
| 130 |
+
"""Color-distance segmentation -> morphology -> contours -> geometric filter."""
|
| 131 |
+
rgb = np.array(image.convert("RGB")).astype(np.int16)
|
| 132 |
+
dist = lambda c: np.sqrt(np.sum((rgb - np.array(c)) ** 2, axis=-1))
|
| 133 |
+
mask = ((dist(BG_COLOR) > 60) & (dist(SHELF_COLOR) > 60)).astype(np.uint8) * 255
|
| 134 |
+
k = np.ones((3, 3), np.uint8)
|
| 135 |
+
mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, k, iterations=2)
|
| 136 |
+
mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, k, iterations=1)
|
| 137 |
+
contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
|
| 138 |
+
|
| 139 |
+
h, w = rgb.shape[:2]
|
| 140 |
+
boxes = []
|
| 141 |
+
for c in contours:
|
| 142 |
+
area = cv2.contourArea(c)
|
| 143 |
+
if not (h * w * min_area_frac <= area <= h * w * max_area_frac):
|
| 144 |
+
continue
|
| 145 |
+
x, y, bw, bh = cv2.boundingRect(c)
|
| 146 |
+
aspect = bw / float(bh) if bh else 0
|
| 147 |
+
if 0.2 <= aspect <= 3.0:
|
| 148 |
+
boxes.append((x, y, bw, bh))
|
| 149 |
+
boxes = _nms(boxes)
|
| 150 |
+
|
| 151 |
+
cell_w = (RACK_W - 2 * MARGIN) / SHELF_COLS
|
| 152 |
+
cell_h = (RACK_H - 2 * MARGIN) / SHELF_ROWS
|
| 153 |
+
cells = set()
|
| 154 |
+
for (x, y, bw, bh) in boxes:
|
| 155 |
+
col = max(0, min(SHELF_COLS - 1, int((x + bw / 2 - MARGIN) // cell_w)))
|
| 156 |
+
row = max(0, min(SHELF_ROWS - 1, int((y + bh / 2 - MARGIN) // cell_h)))
|
| 157 |
+
cells.add((row, col))
|
| 158 |
+
occupancy = len(cells) / (SHELF_ROWS * SHELF_COLS)
|
| 159 |
+
|
| 160 |
+
annotated = image.convert("RGB").copy()
|
| 161 |
+
d = ImageDraw.Draw(annotated)
|
| 162 |
+
for (x, y, bw, bh) in boxes:
|
| 163 |
+
d.rectangle([x, y, x + bw, y + bh], outline=(0, 220, 90), width=3)
|
| 164 |
+
return DetectionResult(len(boxes), float(occupancy), annotated, boxes)
|