RAG in 1 Day: Chat with Your Own Documents

In one day, you'll build a production-grade RAG (Retrieval Augmented Generation) chatbot that lets you "chat with your documents" — PDFs, policies, manuals, contracts. LangChain + ChromaDB + OpenAI + Streamlit. By 5pm, you'll have a working app you can deploy.

What You'll Build

By the end of this workshop, you'll have a working "chat with your PDFs" application:

• 📁 Upload documents (PDF, Word, TXT) via a web UI
• 🧠 Documents are chunked, embedded, and stored in a vector database (ChromaDB)
• 💬 Ask questions in natural language
• 🔎 The system retrieves the most relevant passages
• ✍️ GPT-5/Claude generates a grounded answer with citations
• 🚀 Running locally, ready to deploy to Streamlit Community Cloud or Render

This isn't a toy demo. It's the same architecture I use in production for client RAG systems — just simplified for a one-day build.

Why RAG?

LLMs hallucinate. They'll confidently tell you the wrong answer because they don't know what they don't know. RAG fixes this by grounding the LLM's response in your specific documents. Instead of asking "What's our refund policy?" and getting a generic answer, the system:

1. Searches your actual policy document for "refund"
2. Retrieves the relevant passages
3. Feeds those passages to the LLM as context
4. Generates an answer based only on your document
5. Cites the source passage so you can verify

This is the architecture behind every "chat with your docs" product — from Chatbase to Intercom Fin to custom enterprise systems. And it's learnable in one day.

Prerequisites

You don't need an ML background. You do need:

• ✅ Basic Python — functions, pip, virtual environments
• ✅ Understanding of APIs and JSON — you've called a REST API before
• ✅ A laptop with Python 3.10+ installed
• ✅ An OpenAI API key (or Anthropic key — we support both)
• ❌ No ML/AI background needed — we teach everything from scratch

Tools We'll Use

Workshop Curriculum

Morning: Foundations (9:00 AM – 1:00 PM)

Module 1: Intro & Why RAG (9:00 – 9:45)

• Why LLMs hallucinate and how RAG fixes it
• RAG vs. fine-tuning vs. prompt engineering — when to use each
• Real-world RAG examples (Chatbase, Intercom Fin, enterprise knowledge bases)
• Workshop overview and what you'll build

Module 2: Embeddings & Vector Math (9:45 – 10:30)

• What are embeddings? (text → vectors of numbers)
• Cosine similarity — how "closeness" is measured
• Hands-on: embed sentences, calculate similarity, see it work
• Choosing an embedding model (OpenAI text-embedding-3-large, alternatives)

Module 3: Chunking Strategies (10:45 – 11:30)

• The #1 RAG failure: bad chunking
• RecursiveCharacterTextSplitter — chunk size and overlap
• Semantic chunking — splitting by meaning, not character count
• Hands-on: chunk a real PDF, inspect the results, compare strategies

Module 4: Vector Databases with Chroma (11:30 – 12:30)

• What a vector database does (and why you need one)
• ChromaDB — local, free, zero-setup
• Hands-on: build a vector store, add documents, similarity search
• Metadata filters — search by document, date, category
• When to upgrade to Pinecone, Qdrant, or Weaviate

Q&A + Lunch (12:30 – 1:00)

Afternoon: Build & Ship (2:00 PM – 6:00 PM)

Module 5: LangChain RAG Chain (2:00 – 3:00)

• LangChain architecture: retriever + chain + prompt
• Building the RAG chain with create_retrieval_chain
• System prompts — controlling tone, format, and grounding behavior
• Hands-on: wire up the full chain, ask your first question

Module 6: Streamlit Chat UI (3:00 – 3:45)

• Streamlit basics — chat_message, chat_input, session state
• Building a chat interface like ChatGPT
• Adding source citations (show which passage the answer came from)
• File upload widget — let users upload their own PDFs
• Hands-on: ship the full UI

Module 7: Production Hardening (4:00 – 5:00)

• Evaluation: RAGAS (faithfulness, answer relevance, context recall)
• Re-ranking — improving retrieval quality with a cross-encoder
• Hybrid search — combining keyword + semantic search
• Guardrails — preventing off-topic questions, PII leakage
• Cost and latency optimization — caching, model selection
• When to use local models (Ollama) vs. cloud APIs

Module 8: Deploy & Wrap-Up (5:00 – 6:00)

• Deploy to Streamlit Community Cloud (free)
• Alternative: Render, Railway, or Docker + VPS
• Adding authentication (Streamlit Authenticator)
• Q&A and troubleshooting
• Certificate of completion

What's Included

• ✅ Full-day live instruction (hands-on, not lecture-style)
• ✅ Private code repository — starter code + finished solution
• ✅ Slide deck + cheat sheets (yours to keep)
• ✅ Curated practice dataset (sample PDFs to work with)
• ✅ Certificate of completion
• ✅ 30-day post-workshop support — Slack/Discord channel + 1 office-hours call
• ✅ Recording of all sessions (in case you miss a module or want to review)

Pricing

• Team discounts: 3+ from the same company — 15% off each
• Scholarships: Partial scholarships available for African-based builders (apply via email)
• Company sponsorship: Invoice provided for company-sponsored seats

Format & Next Dates

• Format: Remote (Zoom + live coding) or in-person (Nairobi)
• Cohort size: Capped at 20 for hands-on help
• Frequency: Monthly cohorts
• Next dates: To be announced — join the waitlist below for early notification

FAQ