RAG System

中文 | English

Retrieval-Augmented Generation system for documents, built in Go with a Connect/gRPC API and a Next.js (Bun) frontend.

Features

• Document pipeline: Presigned upload → PDF text extraction (Doc2X) → semantic chunking → embeddings → pgvector storage
• Hybrid retrieval: Vector search + keyword-aware rerank, then LLM summarization
• Clients: Connect/gRPC; generated Go/TypeScript stubs; Bun/Next.js demo UI with virtual scrolling
• Infra: PostgreSQL + pgvector, Redis cache, MinIO object storage
• Ops: just gen for proto codegen via Docker, just web-* for frontend tasks

Architecture

The system follows a modular architecture with clean separation of concerns:

cmd/server/          # Application entry point
internal/
├── adapters/        # Database adapters (PostgreSQL)
├── clients/         # External service clients
│   ├── base/        # Shared HTTP client utilities
│   ├── doc2x/       # Document parsing service
│   ├── embedding/   # Embedding generation
│   ├── openai/      # LLM integration
│   └── rerank/      # Document reranking
├── config/          # Configuration management
├── logger/          # Structured logging
├── redis/           # Cache layer
├── server/          # HTTP server and handlers
└── storage/         # File storage (MinIO)

Requirements

• Go 1.21+ (recommended)
• PostgreSQL with pgvector
• Redis
• MinIO (or S3-compatible) for object storage
• Docker (for just gen codegen) and Bun (for the Next.js frontend)

Quickstart

1. Clone & Go deps

git clone https://github.com/hsn0918/rag.git
cd rag
go mod download

2. Configure: copy config.example.yaml to config.yaml and fill in DB/Redis/MinIO/API keys.

3. Generate code (requires Docker):

just gen

4. Run server:

go run cmd/server/main.go

5. Frontend (Bun + Next.js):

cd web
bun install
bun run dev

Configuration

Use config.yaml (see config.example.yaml) or env vars:

• server: host/port
• database: PostgreSQL + pgvector DSN parts
• redis: host/port/auth
• minio: endpoint/access keys/bucket
• services: Doc2X, Embedding, Reranker, LLM endpoints + models/API keys
• chunking: chunk sizes/overlap/semantic options

API (Connect/gRPC)

Service: rag.v1.RagService

• POST /rag.v1.RagService/PreUpload — presigned upload URL
• POST /rag.v1.RagService/UploadPdf — process & index PDF
• POST /rag.v1.RagService/GetContext — full RAG pipeline (keywords → embedding → search → rerank → summarize)
• POST /rag.v1.RagService/ListDocuments — cursor-paginated doc list
• POST /rag.v1.RagService/DeleteDocument — delete doc and chunks

See api/rag/v1/rag.proto for message shapes; generated clients in internal/gen (Go) and web/gen (TS).

CLI & Scripts

• just gen — buf codegen (via Docker), fixes TS import extensions
• just web-install|web-dev|web-build|web-start|web-lint — frontend tasks (Bun)
• go run cmd/server/main.go — start backend

Frontend (Next.js + Bun)

• Uses generated Connect TS client; proxy to backend via web/next.config.mjs (/api -> BACKEND_URL).
• Includes upload flow, RAG query, document list (cursor pagination, delete), keyword D3 visualization, and structured answer rendering.

Notes

• For codegen, Docker access is required (bufbuild/buf container).

• For embedding/rerank/LLM services, update API keys and models in config.yaml.

• Ensure pgvector extension exists in Postgres and the DSN matches your config.

• Database: github.com/jackc/pgx/v5 for PostgreSQL

• Vector Search: github.com/pgvector/pgvector-go for embeddings

• Cache: github.com/redis/rueidis for Redis

• Storage: github.com/minio/minio-go/v7 for object storage

• HTTP Client: github.com/go-resty/resty/v2 for external APIs

• Dependency Injection: go.uber.org/fx for application structure

• Logging: log/slog for structured logging

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

License

This project is licensed under the terms specified in the repository.