A learner-focused tutorial that guides developers step-by-step through building a production-grade Agentic RAG research assistant. It prioritizes mastering keyword search fundamentals before adding semantic vectors and uses a multi-container local infrastructure.
Project overview
Teaches the professional path of mastering keyword search foundations before enhancing retrieval with vectors and agentic workflows.
Project type
Learning Resources
Use cases
Search & Research · Learning & Education
Deployment
Refer to project documentation
License
MIT
Best for
Developers, AI engineers, and educators who want to learn how to build production-grade RAG systems step-by-step.
Key capabilities
Complete infrastructure setup with Docker Compose, FastAPI, PostgreSQL, OpenSearch, Airflow, and Ollama local LLM.
Automated data pipeline fetching and parsing academic papers from arXiv using Docling.
Production BM25 keyword search with OpenSearch featuring filtering and relevance scoring.
Intelligent chunking combined with hybrid search merging keywords with semantic understanding using Jina AI embeddings.
RAG pipeline with local LLM via Ollama, streaming responses, and an interactive Gradio web interface.
RAG pipeline monitoring and tracing with Langfuse, plus Redis caching for optimized performance.
Agentic RAG workflow with LangGraph including intelligent decision-making, document grading, query rewriting, and guardrails.
Telegram bot providing mobile access and conversational AI for the research assistant.
Limitations and risks
Requires a minimum of 8GB+ RAM and 20GB+ free disk space.
Requires Docker Desktop, Python 3.12+, and UV Package Manager to be installed and manually configured.
Getting started
Install Docker Desktop, Python 3.12+, and UV Package Manager. Clone the repository, copy .env.example to .env, run uv sync, and start the services using docker compose up --build -d.