Quest

Quest is a Retrieval-Augmented Generation (RAG) engine designed to assist with solving and explaining Data Structures and Algorithms (DSA) problems. It leverages a combination of retrieval-based methods and generative models to provide accurate and context-aware solutions to coding problems, explanations of concepts, and metadata about LeetCode-style questions.

Features

• Exact Matching Retrieval: Quickly retrieves solutions for exact matches of problem titles from a curated dataset.
• Context-Aware Generation: Generates detailed explanations and solutions for general queries using a generative model.
• Metadata Integration: Provides metadata such as problem difficulty, related topics, and edge cases for retrieved solutions.
• Efficient Search: Uses FAISS and HNSW for fast and accurate similarity search in high-dimensional spaces.
• Customizable: Supports different models and configurations for retrieval and generation.
• Pre-Generated Solutions: Includes 1800+ solutions in JSON format, created using the qwen2.5-coder:1.5b model on a local machine.
• Optimized Inference Speed: Reduced query retrieval time from 1 minute to 15 seconds and reasoning model time from 20 minutes to under 4 minutes.
• Dynamic Prompt Switching: Implements dynamic prompt switching for Qwen and Deepseek models to prevent infinite reasoning loops.
• Unknown Query Handling: Switches to model’s own inference for queries with confidence below 0.6 for better accuracy.
• HNSW Indexing: Fine-tuned HNSW parameters for faster and more relevant query retrieval.
• Memory Buffer: Retains past interactions for more contextually accurate responses over multiple queries.
• Dataset Generation: Created an on-device dataset of over 1850 coding questions JSON, enriched with metadata.
• Reasoning Model Integration: Integrated Deepseekr1 7B for complex problem-solving with optimized prompt templates.
• Frontend Interface: Built with Flask and Bootstrap, featuring a clean and responsive design with interactive elements.
• Search and Query Optimization: Optimized query engine with efficient indexing and metadata-based retrieval.
• Code Quality: Well-structured code, easy to understand, extend, and deploy, hosted on GitHub for public access.

---

Installation

1. Clone the repository:

   git clone https://github.com/udit-rawat/Quest.git
   cd Quest

2. Install the required dependencies:

   pip install -r requirements.txt

3. Unzip the leetcode_solutions.zip file to access the solutions in JSON format:

   unzip leetcode_solutions.zip -d src/DSAAssistant/components/

---

Setting Up Ollama

To use the generative models (qwen2.5-coder:1.5b and deepseek-r1:1.5b), you need to install Ollama and pull the required models.

Install Ollama

• macOS: Download and install Ollama from the official website:
  Ollama for macOS

• Windows: Download and install Ollama from the official website:
  Ollama for Windows

Pull the Models

Once Ollama is installed, pull the required models using the following commands:

# Pull qwen2.5-coder:1.5b
ollama pull qwen2.5-coder:1.5b

# Pull deepseek-r1:1.5b
ollama pull deepseek-r1:1.5b

---

Usage

Running the RAG Engine

1. Start the RAG engine:

   python app.py

2. Use the engine to query solutions or explanations. Example:

   Query: "Explain the concept of dynamic programming."

---

Project Structure

Quest/
├── app.py                          # Main application entry point
├── requirements.txt                # List of dependencies
├── .gitignore                      # Files and directories to ignore
├── README.md                       # Project documentation
├── src/
│   ├── DSAAssistant/               # Core components of the RAG engine
│   │   ├── components/
│   │   │   ├── retriever2.py       # LeetCode retriever implementation
│   │   │   ├── rag_engine.py       # RAG engine implementation
│   │   │   ├── leetcode_hnsw.index # Pre-built HNSW index
│   │   │   ├── leetcode_metadata.pkl # Metadata for LeetCode problems
│   │   │   ├── leetcode_solutions.zip # Zipped JSON solutions
│   │   │   └── ...                 # Other components
│   │   └── ...                     # Additional modules
│   └── ...                         # Other source files
└── ...                             # Configuration and other files

---

Configuration

The RAG engine can be configured using the following parameters:

• Retriever:

  • index_path: Path to the HNSW index file.
  • metadata_path: Path to the metadata file.
  • model_name: Name of the sentence transformer model for encoding.

• Generation:

  • ollama_url: URL for the Ollama API (if using a remote model).
  • model_name: Name of the generative model.
  • temperature: Controls the randomness of the generated output.
  • top_p: Controls the diversity of the generated output.

Example configuration:

retriever = LeetCodeRetriever(
    index_path="path/to/leetcode_hnsw.index",
    metadata_path="path/to/leetcode_metadata.pkl",
    model_name="all-MiniLM-L6-v2"
)

rag_engine = RAGEngine(
    retriever=retriever,
    ollama_url="http://localhost:11434/api/generate",
    model_name="qwen2.5-coder:1.5b",
    temperature=0.4,
    top_p=0.9
)

---

Contributing

Contributions are welcome! If you'd like to contribute to Quest, please follow these steps:

1. Fork the repository.
2. Create a new branch for your feature or bugfix:

   git checkout -b feature/your-feature-name

3. Commit your changes:

   git commit -m "Add your feature or fix"

4. Push your branch:

   git push origin feature/your-feature-name

5. Open a pull request and describe your changes.

---

Acknowledgments

• FAISS and HNSW for efficient similarity search.
• Sentence Transformers for encoding text into embeddings.
• Ollama for providing the generative model API.

---

Contact

For questions or feedback, feel free to reach out:

• Udit Rawat
• GitHub: udit-rawat
• Email: [email protected]