🧠 Homework Project: Build a RAG (Retrieval-Augmented Generation) System

📌 Objective

The goal of this project is to build a Retrieval-Augmented Generation (RAG) system that fetches information from external documents and uses it to answer user questions. This project demonstrates how to ground a language model in real-world, up-to-date knowledge.

💡 Note: The selected model has a knowledge cutoff in August 2024. Your system must use retrieved documents to correctly answer questions about events occurring after this date.

---

🛠️ Core Requirements

1. Document Indexing

• Use ChromaDB with persistence enabled.
• Select a document describing an event that happened after August 2024.
• Split the document into at least 50 chunks using appropriate text splitting strategies.

2. System Architecture

• Use the model: gemini-2.0-flash
• Implement the pipeline using:
  • LangChain or LlamaIndex
  • LangSmith or LangFuse for observability and tracing
• Version control your code with Git and GitHub
• Do not use pre-built agents
• Implement:
  • 🗣️ Dialog flow (multi-turn interaction)
  • 🧠 Memory (to track context across messages)

3. Experimentation and Effectiveness Testing

• Create at least 5 meaningful questions that the system should answer using the retrieved document.
• The questions must not be answerable by the language model alone.
• Validate that the system answers correctly only when using retrieval.
• Compare and document the impact of different system prompts on model behavior.

4. Reproducibility & Clean Code Practices

• Use a clean GitHub repository:
  • ❌ No large files in git history
  • ❌ No secret tokens in commit history
• Your code should be:
  • Well-documented
  • Easy to run
  • Clearly structured

---

🚀 Submission Instructions

• Deadline: 11.05 at 23:59
• Each student has a dedicated branch named after them.
• Open a Pull Request (PR) from your working branch to your assigned branch in this repository.
• Your PR must include:
  • ✅ Your full implementation code
  • ✅ A Jupyter notebook or script showing:
    • Document indexing
    • Retrieval steps
    • Question answering
    • Prompt variations and experiments
  • ✅ Link to your LangSmith or LangFuse project

---

⭐ Bonus (Mandatory for Extra Credit)

To earn bonus points, your system must implement both of the following features:

• 🔍 Metadata filtering to refine document retrieval
• 🔁 Multi-Query retrieval (e.g., query rephrasing or multiple simultaneous questions to improve answer quality)

---

Happy building! 🚀