outpaczProduction RAG & Knowledge Systems
Build, evaluate and deploy RAG systems over real documents in 6 weeks: parsing, retrieval, reranking, evals and production. Hands-on, live labs.
27 lessons · 26 hours · ₹12,000
What you will build
Design a RAG architecture
Choose between RAG, fine-tuning and long context, and pick the right stack for the problem
Ingest messy documents
Parse PDFs, tables, scans and Indic-language documents into clean, well-chunked data with metadata
Raise retrieval quality
Use hybrid search, reranking and query rewriting, and prove the gains with recall and MRR
Generate grounded answers
Produce cited answers, route between documents and SQL, and support Hinglish queries
Evaluate and secure the system
Run golden-set evals with Ragas, trace failures, and defend against prompt injection and data leaks
Deploy to production
Ship a FastAPI service with caching, re-indexing, and cost and latency budgets
Curriculum
Bridge videos (prerequisites)
- Python for LLM work · 180 min
- Prompt writing basics · 90 min
- SQL and DBMS basics · 180 min
Week 1: RAG foundations
- RAG architecture end to end · 20 min
- RAG vs fine-tuning vs long context · 15 min
- Embeddings and chunking in plain terms, and stack selection · 25 min
- Live lab: your first RAG on a real PDF set · 120 min
Week 2: Data ingestion
- Parsing PDFs, tables, scans and HTML · 20 min
- OCR for Indic scripts and metadata design · 15 min
- Chunking strategies: recursive, semantic, structure-aware, parent-child · 25 min
- Live lab: parse a messy corpus and compare chunking strategies · 120 min
Week 3: Retrieval quality
- Vector databases: Pinecone, pgvector, Qdrant · 20 min
- Hybrid search and reranking · 20 min
- Query rewriting, multi-query and HyDE · 20 min
- Live lab: measure recall@k and MRR on 30 golden questions · 120 min
Week 4: Answer generation
- Grounded prompts, citations and I don't know behaviour · 20 min
- Text-to-SQL and routing between documents and SQL · 20 min
- Multilingual and Hinglish RAG, and when GraphRAG pays off · 20 min
- Live lab: policy assistant with citations and a SQL route · 120 min
Week 5: Evaluation and security
- Golden sets and Ragas metrics · 20 min
- Tracing with Langfuse and LangSmith · 15 min
- Prompt injection, PII and access control · 25 min
- Live lab: find and fix your top failure classes · 120 min
Week 6: Production
- FastAPI, streaming, auth and rate limits · 20 min
- Caching, incremental re-indexing, cost and latency budgets · 20 min
- Docker, deployment and feedback loops · 20 min
- Live: capstone demo day and viva · 120 min
Syllabus
Who this course is for
Students and working professionals who want to build AI that answers from their own documents and data, and run it reliably in production.
Prerequisites
Python, basic DBMS and prompt writing. The bridge videos at the start cover all three, about 7.5 hours in total.
How the course runs
- Recorded lessons are released every Monday, in short videos of about an hour a week
- One live session each weekend: a 2-hour hands-on lab with mentors, plus doubt solving. Replays are available
- A problem statement every week, reviewed by mentors, with a leaderboard
- Batch community group with daily doubt-desk hours
- Live session timings are announced before each batch starts
Time commitment
About 18 hours of core sessions over 6 weeks, plus 7.5 hours of bridge videos.
Assessment and certificate
- Weekly problem statements
- Capstone: a deployed RAG API with a UI and an evaluation report
- A short mentor viva
- Verified certificate on scoring 70% or above with the capstone submitted
