Update README.md
Browse files
README.md
CHANGED
|
@@ -7,37 +7,31 @@ sdk: static
|
|
| 7 |
pinned: true
|
| 8 |
license: apache-2.0
|
| 9 |
---
|
| 10 |
-
#
|
| 11 |
|
| 12 |
-
**
|
|
|
|
|
|
|
| 13 |
|
| 14 |
-
|
| 15 |
-
and *govern* production AI agents on top of the systems they already have.
|
| 16 |
-
The platform is LangGraph-native and pairs agent building with a real
|
| 17 |
-
governance layer — observability, policy enforcement, ROI measurement, and
|
| 18 |
-
EU AI Act compliance.
|
| 19 |
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
|
| 24 |
-
|
|
|
|
|
|
|
|
|
|
| 25 |
|
|
|
|
|
|
|
|
|
|
| 26 |
|
| 27 |
-
|
|
|
|
|
|
|
|
|
|
| 28 |
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
**Contract Red-Flag Explainer (caveat)**: on-device model that spots and explains risky clauses in plain language ("what am I actually signing?"). Runs privately in English, Romanian, Polish, and Hungarian, returns structured JSON across a 34-clause taxonomy. Qwen3-4B fine-tune, benchmarked on RedFlag-Bench.
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
## Coming soon
|
| 36 |
-
|
| 37 |
-
- **Ontology Model** — generates and maps domain ontologies for agent grounding.
|
| 38 |
-
- **BPMN Generation** *(planned)* — natural language → BPMN process models.
|
| 39 |
-
- **UI Generation** *(planned)* — interfaces generated on the fly.
|
| 40 |
-
|
| 41 |
-
**Follow us** to get these the day they drop.
|
| 42 |
-
|
| 43 |
-
[flowx.ai](https://flowx.ai) · [github.com/flowx-ai](https://github.com/flowx-ai) · [LinkedIn](https://linkedin.com/company/flowxai)
|
|
|
|
| 7 |
pinned: true
|
| 8 |
license: apache-2.0
|
| 9 |
---
|
| 10 |
+
# FlowX.AI
|
| 11 |
|
| 12 |
+
**Open, on-device AI for regulated industries.** We build small, private, fast models that run
|
| 13 |
+
on-premise with **zero data egress**, so banks, insurers, and logistics operators can use AI
|
| 14 |
+
without sending customer data to a cloud.
|
| 15 |
|
| 16 |
+
## 🧠 OpenNER — small on-device extraction models
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
+
28 fine-tuned encoder models for structured extraction, PII detection, and de-identification.
|
| 19 |
+
**Apache-2.0.** Benchmarked against the latest frontier LLMs (GPT-5.4, Gemini 3.x, Claude Haiku):
|
| 20 |
+
matching or beating them on structured extraction at **~100 ms/doc, $0, and zero data egress**.
|
| 21 |
|
| 22 |
+
- 🏦 **[FlowX Banking Models](https://huggingface.co/collections/flowxai/flowx-banking-models-6a5c9be270b1ab8d0c02566d)** — IBAN/PAN/KYC/AML + SEC-filing facts (8 models)
|
| 23 |
+
- 🛡️ **[FlowX Insurance Models](https://huggingface.co/collections/flowxai/flowx-insurance-models-6a5c9be2ed35b4a126afe56d)** — policy/claims/VIN + GLBA·HIPAA de-id (8 models)
|
| 24 |
+
- 🚢 **[FlowX Logistics Models](https://huggingface.co/collections/flowxai/flowx-logistics-models-6a5c9be213707fd7f44728ab)** — container/port/hazmat/bill-of-lading + trade de-id (7 models)
|
| 25 |
+
- 🌐 **[FlowX Cross-Industry Models](https://huggingface.co/collections/flowxai/flowx-cross-industry-models-6a5c9be38707b1635019b044)** — multilingual PII (9 languages), privacy filter, invoice/CV/KYB (5 models)
|
| 26 |
|
| 27 |
+
Every identifier is **checksum-validated** (IBAN mod-97, card Luhn, ISIN/LEI, container ISO-6346,
|
| 28 |
+
VIN, 8 national IDs). Ships with a de-identification engine (6 policy profiles + signed audit
|
| 29 |
+
reports) and ONNX / INT8 edge exports. Runs on CPU, Apple Silicon, one GPU, or in the browser.
|
| 30 |
|
| 31 |
+
## Other model families
|
| 32 |
+
- **Semantic Mapper · Sentinel Gate** — ontology extraction + escalation gating
|
| 33 |
+
- **Scam Guard** — scam / fraud detection
|
| 34 |
+
- **BPMN Gen** — business-process generation
|
| 35 |
|
| 36 |
+
---
|
| 37 |
+
*Small models, on your hardware. Apache-2.0. [flowx.ai](https://flowx.ai)*
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|