# SafeVixAI — Research Notebooks 📓

These notebooks are the **research and training layer** of SafeVixAI. Each one processes raw data from the Hub and produces a model, index, or processed dataset used by the live application.

---

## ⚡ One-Click Colab Setup

Run this at the top of **any** notebook to mount the full dataset:

```python
# Step 1 — Clone the Hub (only run once per session)
!git clone https://huggingface.co/datasets/SafeVixAI/SafeVixAI-Dataset-Hub /content/hub

# Step 2 — Install dependencies
!pip install -q pdfplumber chromadb sentence-transformers ultralytics onnx

# Step 3 — Confirm data is accessible
import os
print("Data folders:", os.listdir("/content/hub"))
```

---

## 📓 Notebook Index

| # | Notebook | What it Produces | Input Data |
|---|---|---|---|
| 1 | `YOLOv8_Pothole_Detector_Training` | ONNX road damage model | `pothole_training/road_damage_2025/` |
| 2 | `ChromaDB_RAG_Vectorstore_Build` | ChromaDB index for legal RAG | `legal/`, `medical/` PDFs |
| 3 | `Accident_EDA_&_Hotspot_Generator` | Blackspot seed CSV + heatmap | `accidents/kaggle_india_accidents.csv` |
| 4 | `Roads_Data_Processing` | Sampled PMGSY GeoJSON | `roads/pmgsy_roads.geojson` |
| 5 | `Risk_Model_ONNX_Training` | Risk scoring ONNX model | `accidents/` + `roads/` |

---

## 📁 Data Paths Used by These Notebooks

All notebooks expect data at these paths after cloning the Hub:

```
/content/hub/
├── chatbot_service/data/
│   ├── accidents/              ← Notebooks 3, 5
│   ├── legal/                  ← Notebook 2
│   ├── medical/                ← Notebook 2
│   ├── pothole_training/       ← Notebook 1
│   └── roads/                  ← Notebook 4
├── backend/datasets/           ← Notebook 4, 5
└── frontend/public/models/     ← Output for Notebooks 1, 5
```

---

## 🏃 Running Order

For a full pipeline run, execute notebooks in this order:

```
1 → Train pothole detector → produces ONNX model
2 → Build RAG vectorstore  → produces ChromaDB index
3 → Accident EDA           → produces blackspot_seed.csv
4 → Roads processing       → produces sampled GeoJSON
5 → Risk model             → combines accidents + roads → risk ONNX
```

---

## 💾 Saving Outputs Back

After training, outputs land in `/content/hub/`. To persist them:

```python
# Commit outputs back to Hub (requires HF token)
import os
os.environ["HUGGING_FACE_TOKEN"] = "your_token_here"

!cd /content/hub && git config user.email "you@example.com"
!cd /content/hub && git config user.name "SafeVixAI"
!cd /content/hub && git add . && git commit -m "chore: update trained model outputs"
!cd /content/hub && git push https://your_username:$HUGGING_FACE_TOKEN@huggingface.co/datasets/SafeVixAI/SafeVixAI-Dataset-Hub main
```

---

*Part of the [SafeVixAI](https://github.com/SafeVixAI/SafeVixAI) — IIT Madras Road Safety Hackathon 2026*