Rwanda-AI-Network/SomaAI

Open-source AI platform for Rwanda education # SomaAI RAG-powered educational assistant for Rwandan students and teachers. Transforms curriculum PDFs into searchable knowledge and generates grounded, cited answers. **Stack**: FastAPI · Qdrant (384d) · PostgreSQL 16 · Redis 7 · Groq (Llama 3.2) --- ## Architecture ```mermaid graph LR Client["Client"] -->|POST /chat/ask| API["FastAPI Gateway"] API --> RAG["RAG Pipeline"] RAG --> Q["Qdrant Vectors"] RAG --> LLM["Groq Llama 3.2"] RAG --> Cache["Redis Cache"] API --- PG["PostgreSQL Metadata"] Ingest["Ingestion Pipeline"] --> Q Ingest --> PG ``` **How it works**: Curriculum PDFs are ingested through a 7-stage pipeline (extract → chunk → embed → store). Student/teacher queries hit the RAG pipeline, which classifies, retrieves, generates, and validates citations. --- ## Quick Start ```bash git clone github.com cd SomaAI cp .env.example .env # Set GROQ_API_KEY in .env (or LLM_BACKEND=mock with TESTING=1 for dev) uv sync make docker ``` App runs at localhost · Swagger at localhost --- ## Documentation | Document | Description | |----------|-------------| | ARCHITECTURE.md | System design, module breakdown, data flow, request lifecycles | | INGESTION_PIPELINE.md | 7-stage ingestion: extract → chunk → embed → store | | RETRIEVAL.md | Dense retrieval, fallback strategy, reranker/BM25 status | | DEVELOPMENT.md | Local setup, environment variables, debugging, testing | | ROADMAP.md | MVP status, prioritized improvements | | monitoring.md | Prometheus metrics, alerts, and Grafana dashboards | | CONTRIBUTING.md | Branch strategy, commit conventions, PR process | | docs/api.md | API endpoint reference | | CHANGELOG.md | Release history | | CODE_OF_CONDUCT.md | Community guidelines | | SECURITY.md | Security policy | --- ## API Overview | Endpoint | Method | Description | |----------|--------|-------------| | `/api/v1/chat/ask` | POST | Ask a question (returns AI answer + citations) | | `/api/v1/ch …