AI PR Reviewer for PotPie AI

Project Overview

An automated code review tool built with FastAPI, Celery, and Redis, deployed on Render. The application provides intelligent code review capabilities for GitHub pull requests.

🔗 Frontend Deployment: https://ai-code-review-fe.vercel.app/ 🔗 Live API URL : https://fastapi-app-y507.onrender.com

Tech Stack

• Backend: FastAPI
• Task Queue: Celery
• Caching: Redis
• AI Model: Mistral AI
• Frontend: Next.js
• Deployment: Vercel (Frontend), Render (Backend)

Setup Instructions

1. Clone the repository:

   git clone https://github.com/kinshuksinghbist/AI_Code_Review_BE.git

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

3. Install dependencies:

   pip install -r requirements.txt

4. Start the FastAPI server:

   uvicorn api.app:app --reload

5. Start the Celery worker (in a separate terminal):

   celery -A api.celery.app worker --loglevel=info

Known Issues currently ⚠️ In some cases, you may need to retry the same request multiple times to successfully retrieve the review results This can happen due to temporary system constraints If initial request doesn't return results, wait briefly and resubmit the request

Branches

• main: Deployment-ready version (without async Celery)
• local: Development branch with full async Celery support

Design Rationale

• Next.js: Chosen for frontend development and easy deployment
• Redis: Selected for fast caching and personal experience with redis and redis-cli
• Mistral AI & Langchain: Utilized for straightforward AI integration. Also mistral ai is free
• Deployment Strategy: Leveraged Vercel and Render for efficient hosting.

Key Features

• Automated GitHub PR code reviews
• Asynchronous task processing
• Caching of review results using redis
• Flexible AI-powered analysis

Documentation

Full API documentation with curl requests is available at https://fastapi-app-y507.onrender.com/api/docs

API Endpoints

1. Analyze PR Review

• Endpoint: POST /api/analyze-pr
• Purpose: Submit a PR for code review
• Request Body:

  {
    "repo_owner": "string",
    "repo_name": "string",
    "pr_number": 123,
    "github_token": "string"
  }

2. Get Review Status

• Endpoint: GET /api/status/{task_id}
• Purpose: Check the status of a code review task

3. Get Review Results

• Endpoint: GET /api/results/{task_id}
• Purpose: Retrieve the final results of a code review task

Future Improvements

• Define more structured response formats,
• Implement more granular AI task segmentati and go through the git diffs assiging smaller and more precise roles to the ai agent, giving it more powerful tools
• Add Grafana logging
• Add async processing for deployment environments