🤖 DevMentor: AI Software Architecture Assistant

DevMentor is a Retrieval-Augmented Generation (RAG) system designed to act as an AI assistant for developers.
It helps users, especially those new to a project, understand complex software codebases by answering natural-language questions about architecture, code logic, and historical decisions.

---

🧠 Core Concept

Onboarding onto a new, large software project is a major challenge for developers.
Finding out why a certain technology was chosen, how a specific module works, or where to start contributing can take days or weeks of digging through documentation and asking colleagues.

DevMentor solves this by ingesting an entire codebase—including source code, documentation (.md, .pdf, .docx), Jupyter Notebooks, and Architecture Decision Records (ADRs)—and making it conversationally accessible.
It acts as an infinitely patient senior developer, ready to answer your questions and guide you through the project.

---

✨ Key Features (Current Phase)

• Flexible Data Ingestion: Ingests knowledge from either local repositories or directly from a public GitHub URL.
• Multi-Repository Support: The application can manage and query multiple, distinct knowledge bases.
• Smart File Processing: Uses a multi-layered filtering system to automatically ignore irrelevant files and a dispatcher to correctly parse various file types.
• Conversational Q&A: Ask questions about the codebase in plain English and get detailed, context-aware answers.
• Intelligent RAG Pipeline: Built with LangChain, using a FAISS vector store for efficient retrieval and the Google Gemini 2.5 Flash API for powerful generation.
• Persona-Driven AI: The prompt is engineered to make the AI act as a helpful and patient mentor, capable of explaining concepts in a beginner-friendly way.
• Streaming Responses: The user interface displays answers with a real-time "typewriter" effect for a responsive and modern user experience.
• Dockerized Environment: The entire application is containerized using Docker and managed with Docker Compose, ensuring a consistent and reproducible setup for both the web app and a command-line interface.
• Interactive Web UI: A clean and user-friendly web interface built with Streamlit, featuring a full chat history and a knowledge base management page.
• GPU Acceleration: Provides an optional GPU-enabled installation path for faster ingestion and retrieval.

---

🛠️ Tech Stack

Component	Technologies Used
Backend & AI	Python, LangChain, Google Gemini 2.5 Flash
Vector Store	FAISS (CPU & GPU versions)
Embeddings	Hugging Face Sentence Transformers (BAAI/bge-small-en-v1.5)
Frontend	Streamlit
Deployment & Tooling	Docker, Docker Compose, GitPython

---

🚀 Getting Started

Follow these steps to set up and run DevMentor on your local machine.

🧩 Prerequisites

• Git
• Docker
• Docker Compose (usually included with Docker Desktop)
• (Optional) For GPU support: A compatible NVIDIA GPU with the appropriate drivers and CUDA toolkit installed.

---

1️⃣ Clone the Repository

git clone https://github.com/c0mrade03/DevMentor.git
cd DevMentor

2️⃣ Set Up Environment Variables

The application requires a Google API key to function.

1. Create a file named .env in the root of the project.
2. Copy the contents of .env.example into your new .env file.
3. Add your Google Gemini API key to the .env file.

Example .env file:

GOOGLE_API_KEY="your_google_api_key_here"

3️⃣ Installation

This project supports both CPU-only and GPU-accelerated environments.

🖥️ Default Installation (CPU-only)

1. Create and activate a Python virtual environment:

   python3 -m venv venv_cpu
   source venv_cpu/bin/activate
   

2. Insall all required packages:

   pip install -r requirements.txt
   

⚡ Optional: GPU Acceleration

1. Create and activate a Python virtual environment:

   python3 -m venv venv_gpu
   source venv_gpu/bin/activate
   

2. Install the GPU-specific packages:

   pip install -r requirements_gpu.txt
   

4️⃣ Build a Knowledge Base

You must ingest a repository before you can ask questions.

Option A: From a GitHub URL (Recommended)

python -m ingest.create_vectorstore --url https://github.com/cookiecutter/cookiecutter

Option B: From a Local Path (Fallback)

If no URL is provided, the script will default to the path specified in ingest/config.py.

python -m ingest.create_vectorstore

5️⃣ Build the Docker Images

Build the necessary Docker images using Docker Compose:

docker compose build

🖥️ Usage

You can run DevMentor as either a Streamlit web app or a command-line tool.

Running the Web Application

CPU Version

docker compose up app-cpu

Once the container is running, open your web browser and navigate to:
http://localhost:8501 (CPU version)

GPU Version

docker compose up app-gpu

For GPU, navigate to: http://localhost:8502

Running the Command-Line Interface (CLI)

CPU Version

docker compose run --rm cli-cpu --repo <repo_name>

GPU Version

docker compose run --rm cli-gpu --repo <repo_name>

Replace <repo_name> with the name of a folder inside data/vector_stores/.

---

🔮 Future Scope

• Agentic Capabilities (MCP): Give the AI "tools" to perform live actions, such as interacting with the GitHub API to find "Good First Issues" for new contributors.
• Conversational Memory: Implement an explicit memory module to allow for more natural, multi-turn follow-up conversations.
• Source Citing: Add a "Show Sources" feature in the UI to display the exact document chunks used to generate an answer.
• CI/CD Pipeline: Implement a full GitHub Actions workflow for automated testing and linting.

---

📜 License

This project is licensed under the MIT License.
See the LICENSE file for details.