ce-workflow

Multi-agent QA/QC workflow for cold electronics detector components. A LangGraph pipeline (Monitor → DAQ → QC Analysis → Diagnostic → Catalog) acquires ADC waveforms, flags per-channel anomalies, retrieves grounding context from a local RAG knowledge base, and writes a markdown report — all streamed live to the operator UI.

Stack

• Backend — FastAPI, LangGraph, Qdrant (hybrid dense + sparse RAG), Ollama (local LLM + embeddings), HDF5 waveform storage
• Frontend — Vue 3 + Vuetify, Vite

Quickstart

Prerequisites: Docker, uv, Ollama, Node 20+.

# 1. Vector DB
cd docker && docker compose up -d        # Qdrant on :6333

# 2. Local models (one-time)
ollama pull qwen3.5:4b
ollama pull nomic-embed-text:v1.5

# 3. Backend
cd ../backend
cp .env.example .env
uv sync
uv run uvicorn main:app                  # http://localhost:8000

# 4. Frontend (separate terminal)
cd ../frontend
npm install
npm run dev                              # http://localhost:5173

Open the UI, upload a document via the left panel (a sample is at backend/seeds/dune_ce_intro.txt), then click QC Start to run the pipeline.

Tests

cd backend && uv run pytest              # Qdrant + Ollama required for integration tests
cd frontend && npm test

More

• backend/README.md — detailed backend setup, dev auto-reload, seeding
• backend/CONTEXT.md — agent pipeline, SSE event schema, key files
• frontend/CONTEXT.md — Vue component map, shared state, SSE consumer
• CONTEXT-MAP.md — domain entry point
• Issue tracker — GitHub Issues on czczc/ce-workflow