IntelFactor Inference Engine

Edge-native manufacturing quality inspection + root cause analysis.

Runs entirely on NVIDIA Jetson — no cloud required. Built for Edge. Detects defects, explains why they happened, recommends what to do about it, and learns from operator feedback.

Quick Start

Docker (Recommended)

# Clone
git clone https://github.com/tonesgainz/intelfactor-inference.git
cd intelfactor-inference

# Deploy edge-only (no cloud)
./scripts/deploy.sh edge-only

# Or deploy hybrid (with cloud sync to app.intelfactor.ai)
./scripts/deploy.sh hybrid

# Dashboard: http://localhost:8080

Using Make

make build          # Build Docker images
make deploy-edge    # Deploy edge-only mode
make deploy-hybrid  # Deploy hybrid mode
make test           # Run tests
make help           # Show all commands

Manual Install

pip install -e ".[jetson]"
intelfactor-station doctor --config configs/station.yaml
intelfactor-station run --config configs/station.yaml

Architecture

Camera ──► Vision (YOLOv8/TensorRT) ──► DefectIQ Rules ──► RCA Pipeline
                                                              │
                ┌─────────────────────────────────────────────┘
                ▼
    ┌──── Accumulator (SQLite, 30-day rolling, z-score anomaly)
    ├──── Correlator (process parameters, Pearson drift detection)
    ├──── Explainer (Qwen 3B via llama.cpp, bilingual CN/EN)
    └──── Recommender (SOP-linked actions, causal triples)
                │
                ▼
    Operator Dashboard (Chinese-primary, accept/reject feedback)
                │
                ▼
    Causal Triple Store (defect → cause → outcome, verified by operators)

Deployment Modes

Mode	Description	Cloud Dependency
Edge-Only	Everything local, no internet required	None
Hybrid	Local + sync to app.intelfactor.ai	Optional (outbound only)

Edge-Only Mode

cd deploy/edge-only
cp .env.example .env
# Edit .env with your STATION_ID, CAMERA_URI
docker compose up -d

Hybrid Mode (Cloud Integration)

cd deploy/hybrid
cp .env.example .env
# Edit .env with:
#   - CLOUD_API_URL=https://api.intelfactor.ai
#   - CLOUD_API_KEY=your_key_from_app.intelfactor.ai
docker compose up -d

See docs/INTEGRATION.md for full cloud integration guide.

Project Structure

packages/inference/
├── providers/              # GPU-agnostic vision + language providers
│   ├── resolver.py         #   Auto-detect hardware, select optimal models
│   ├── vision_trt.py       #   TensorRT inference (YOLOv8)
│   └── language_llama.py   #   llama.cpp inference (Qwen 3B)
├── rca/                    # 4-Layer Root Cause Analysis
│   ├── accumulator.py      #   Defect pattern accumulation + anomaly
│   ├── correlator.py       #   Process parameter correlation + drift
│   ├── explainer.py        #   Bilingual explanation via SLM
│   └── recommender.py      #   SOP-linked recommendations
├── storage/                # Storage abstraction layer
│   ├── factory.py          #   STORAGE_MODE routing (local/cloud)
│   ├── sqlite_events.py    #   SQLite event store
│   └── sqlite_evidence.py  #   Evidence index + filesystem
├── api_v2.py               # REST API with evidence endpoints
├── sync_cloud.py           # Cloud sync agent
├── evidence.py             # JPEG evidence writer with disk quota
└── static/index.html       # Operator dashboard (Chinese-primary)

deploy/
├── edge-only/              # Single-device deployment
├── hybrid/                 # Edge + cloud sync
└── hub/                    # Site hub (Postgres + Grafana)

docs/
├── ARCHITECTURE.md         # Technical architecture
├── INTEGRATION.md          # Cloud integration guide
└── LOCAL_MODE.md           # Local deployment guide

Hardware Support

Device	VRAM	Role	Vision Model	Language Model
Orin Nano Super	8GB	Station	YOLOv8n FP16	Qwen-2.5-3B Q4
Orin NX 16GB	16GB	Station	YOLOv8m FP16	Qwen-2.5-7B Q4
AGX Orin 64GB	64GB	Hub	YOLOv8l FP16	Qwen-2.5-14B Q8
GPU Server	24GB+	Site hub	YOLOv8x FP16	Qwen-2.5-14B vLLM

Auto-detected via CapabilityResolver.

API Endpoints

Endpoint	Method	Description
/health	GET	Health check
/api/events	GET/POST	Defect events
/api/v1/evidence/:id/image.jpg	GET	Evidence image
/api/triples	GET/PATCH	Causal triples
/api/recommendations	GET	Pending recommendations
/api/feedback	POST	Operator feedback

Full API reference: docs/ARCHITECTURE.md

Tests

# Create venv and run tests
python3 -m venv .venv && source .venv/bin/activate
pip install pytest flask numpy
python -m pytest tests/ -v
# 19 passed

Documentation

• Technical Architecture — System design, data flow, scaling
• Cloud Integration — Connecting to app.intelfactor.ai
• Local Mode — Edge-only deployment guide
• Deployment Guide — Bilingual CN/EN instructions

License

Proprietary — IntelFactor.ai