Project: Knowledge Assistant¶

Building a production-ready RAG (Retrieval-Augmented Generation) application from scratch.

This repository is developed incrementally over 4 weeks, with each branch representing a major milestone in the journey.

🛤️ Journey Philosophy¶

This repository is intentionally built in small, reviewable Pull Requests.

Each Pull Request introduces exactly one major concept, making it easy for anyone to:

Follow the learning journey
Review architectural decisions
Understand how a production RAG system evolves over time
Recreate the project from scratch

The goal isn't just to build an AI application—it's to document the engineering process behind it.

🚀 Learning Roadmap¶

🌱 Week 1 — Core RAG¶

Related project notes: Week 1 AI Assistant

Related theory: Chunking | Embedding | Vector Database | Retrieval | Genration Prams

Branch: week-1-core-RAG

Goal: Build a fully functional RAG pipeline capable of answering questions from uploaded PDF documents.

Progress¶

✅ PR-01 — Project Bootstrap
✅ PR-02 — PDF Upload & Loading
✅ PR-03 — Chunking Engine
✅ PR-04 — Embedding Pipeline
✅ PR-05 — FAISS Vector Store
✅ PR-06 — Basic Question Answering

Outcome¶

By the end of Week 1, the application will:

📄 Upload one or more PDF documents
✂️ Split documents into chunks
🧠 Generate embeddings
🗄️ Store embeddings in FAISS
🔍 Retrieve relevant chunks
💬 Answer user questions using the uploaded documents

🚀 Week 2 — Better Retrieval¶

Related project notes: Week 2 AI Assistant

Related theory: Hybrid Search | Metadata | Guardrails | Structured Outputs

Branch: week-2-better-retrieval

Goal: Improve retrieval quality to make the assistant more production-ready.

Progress¶

✅ PR-07 — Hybrid Search
✅ PR-08 — Metadata Filtering
✅ PR-09 — Guardrails

Outcome¶

The assistant will now support:

🔀 Hybrid Search (Dense + Sparse Retrieval)
🏷️ Metadata-based Filtering
🛡️ Prompt Injection Protection
✅ Safer and more accurate responses

🏗️ Week 3 — LCEL Refactor¶

Related project notes: Week 3 AI Assistant

Related theory: LCEL | Metadata | Guardrails | Structured Outputs

Branch: week-3-LCEL-refactor

Goal: Refactor the application using LangChain Expression Language (LCEL).

Progress¶

✅ PR-10 — LCEL Pipeline
✅ PR-11 — Prompt Refactoring

Outcome¶

The project will evolve from helper functions to a modular pipeline:

Retriever
      │
      ▼
Prompt
      │
      ▼
LLM
      │
      ▼
Output Parser

Benefits:

Better architecture
Easier debugging
Improved composability
Production-style LangChain implementation

🏁 Week 4 — Production Readiness¶

Related project notes: Week 3 AI Assistant

Branch: week-4-production-readiness

Goal: Transform the prototype into a production-quality application.

Progress¶

✅ PR-12 — Chat History
✅ PR-13 — Conversation Memory
✅ PR-14 — Logging & Configuration
✅ PR-15 — Error Handling
✅ PR-16 — Documentation & Deployment

Outcome¶

The final application will include:

💬 Chat History
🧠 Conversation Memory
📊 Logging
⚙️ Configuration Management
🚨 Robust Error Handling
📖 Complete Documentation
☁️ Deployment

📈 Project Evolution¶

Week 1
──────────────
PDF Upload
      │
      ▼
Chunking
      │
      ▼
Embeddings
      │
      ▼
FAISS
      │
      ▼
Question Answering

                ↓

Week 2
────────────────────────
Hybrid Search
      │
      ▼
Metadata Filtering
      │
      ▼
Guardrails

                ↓

Week 3
────────────────────────
LCEL Refactor
      │
      ▼
Cleaner Architecture
      │
      ▼
Better Prompt Pipeline

                ↓

Week 4
────────────────────────
Chat History
      │
      ▼
Conversation Memory
      │
      ▼
Logging
      │
      ▼
Configuration
      │
      ▼
Error Handling
      │
      ▼
Deployment

🎯 Final Features¶

✅ PDF Upload
✅ Recursive Chunking
✅ Embeddings
✅ FAISS Vector Store
✅ Semantic Search
✅ Hybrid Search
✅ Metadata Filtering
✅ Guardrails
✅ LCEL Architecture
✅ Chat History
✅ Conversation Memory
✅ Logging
✅ Configuration Management
✅ Error Handling
⬜ Deployment

📚 Learning Objectives¶

This project is designed to teach:

Retrieval-Augmented Generation (RAG)
Vector Databases
Embeddings
Similarity Search
Hybrid Search
Metadata Filtering
Guardrails
LangChain Expression Language (LCEL)
Production-ready AI application architecture

Every Pull Request represents a single engineering milestone, making it easy to follow the evolution of the project from a minimal RAG prototype to a production-ready AI application.