---

💡 AI DSA Assistant

Flow Diagram

Demo Video

A RAG-based DSA code generator using LangChain, Gemini/Ollama, and FAISS with hybrid search (BM25 + FAISS Ensemble). Uploads DSA PDFs, indexes them, and generates structured solutions (Brute Force → Improved → Optimal).

---

📹 Demo Video

Watch the demo video here

---

🛠️ Tech Stack

• Python 3.10+
• Streamlit - Web UI
• LangChain - RAG framework
• Ollama (Qwen / Mistral / Llama) - Local LLM inference
• Google Gemini API (optional) - Cloud LLM support
• FAISS - Vector database for semantic search
• BM25 (rank_bm25) - Keyword-based search
• Ensemble Retriever - Hybrid search combining BM25 and FAISS
• Automated Evaluation Framework - Retrieval & Generation scoring pipeline

---

🚀 Features

• Hybrid Search: Combines keyword-based (BM25) and semantic-based (FAISS) retrieval for improved accuracy
• Multi-Language Support: Supports C++, Java, and Python DSA problems
• Structured Solutions: Generates three approaches - Brute Force, Improved, and Optimal
• RAG Pipeline: Uses relevant context from DSA PDFs to generate grounded solutions
• Local LLM Support: Fully offline inference using Ollama models
• Automated Evaluation: Built-in benchmarking system to measure retrieval and generation quality
• Streamlit UI: Clean and intuitive web interface

---

📊 Evaluation Framework (NEW)

This project includes a fully automated benchmarking pipeline to measure real RAG performance instead of manual inspection.

Evaluation Stages

1️⃣ Retrieval Evaluation

• Dataset of 50+ DSA problems
• Hybrid retriever returns top-k context
• Checks if important algorithmic keywords exist in retrieved chunks
• Produces Retrieval Hit Rate

Result: 📌 Retrieval Accuracy ≈ 79%

---

2️⃣ Generation Evaluation A separate judge LLM evaluates generated answers on:

• Relevance to the question
• Faithfulness to retrieved context
• Completeness of explanation
• Clarity of response

Each scored from 1 → 5

Average Scores

Metric	Score
Relevance	4.16 / 5
Faithfulness	4.17 / 5
Completeness	4.03 / 5
Clarity	4.19 / 5

📌 Overall grounded response quality ≈ 78%

---

Evaluation Workflow

Dataset → Retriever → Context → Student LLM → Generated Answer → Judge LLM → Scoring → Report

This allows repeatable and objective RAG benchmarking.

---

▶️ Run Locally

Prerequisites

• Python 3.10 or higher
• Ollama installed (recommended)
• (Optional) Google Gemini API key

Installation

git clone https://github.com/Devamsingh09/AI-DSA-Assistant.git
cd AI-DSA-Assistant
pip install -r requirements.txt

(Optional — Gemini)

GOOGLE_API_KEY=your_api_key_here

Run App

streamlit run app/main.py

The app will be available at http://localhost:8501

---

📁 Project Structure

AI-DSA-Assistant/
├── app/
│   ├── indexer.py        # Creates FAISS and BM25 indexes
│   ├── rag_engine.py     # Hybrid search implementation
│   ├── setup.py          # Configuration and constants
│   └── main.py           # Streamlit UI
│
├── evaluator/
│   ├── dataset.json      # 50+ DSA benchmark questions
│   ├── retrieval_eval.py # Retrieval quality scoring
│   ├── generation_eval.py# LLM judge scoring
│   └── run_all.py        # Complete evaluation pipeline
│
├── data/pdfs/            # DSA documents
├── faiss_indexes/        # Generated vector indexes
├── requirements.txt
└── README.md

---

🔧 How It Works

1. Indexing

   • PDFs split into chunks
   • Stored in FAISS (semantic) + BM25 (keyword)

2. Hybrid Retrieval

   • Ensemble Retriever combines both methods

3. Context Generation

   • Top relevant chunks retrieved

4. Code Generation

   • Local LLM generates structured solution

5. Evaluation (Research Mode)

   • Retriever accuracy measured
   • LLM judged by another LLM
   • Produces automated performance report

---

🤝 Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add some amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

---

🙏 Acknowledgments

• LangChain for the RAG framework
• Google for the Gemini API
• Ollama for local LLM inference
• FAISS for efficient vector search
• Streamlit for the web interface

---