CodeSight · AI Code Review Agent 🔍

An autonomous AI agent that clones a GitHub repository, parses source files using AST, and generates structured, confidence-rated code review comments — all through a sleek Streamlit dashboard.

---

Project Overview

CodeSight is a fully agentic AI pipeline for automated code review. Paste any public GitHub URL and the agent:

1. Clones the repository locally using GitPython
2. Parses every Python file using Python's ast module — extracting functions, async functions, and classes as semantic chunks
3. Reviews all chunks in parallel via the Groq LLM API (llama-3.3-70b-versatile)
4. Generates structured JSON review comments with:
   • severity ratings
   • confidence scores
   • actionable suggestions
5. Creates a repository-level health summary:
   • health score (0–100)
   • issue breakdown
   • recurring patterns
   • critical findings
   • recommended next steps
6. Displays everything in an interactive Streamlit dashboard with filters and JSON export

Every issue includes a confidence score (0–100%).

Low-confidence findings are automatically separated under a “Verify This” label — demonstrating responsible AI behavior instead of presenting all findings with equal certainty.

---

Live Demo

https://codesight-ai-code-review-agent.streamlit.app

---

Tech Stack

Layer	Technology
Frontend	Streamlit
Backend	Python
Repository Cloning	GitPython
Parsing Engine	Python AST
LLM Provider	Groq
Model	llama-3.3-70b-versatile
Concurrency	ThreadPoolExecutor
Environment Variables	python-dotenv
Deployment	Streamlit Cloud

---

Setup Instructions

Prerequisites

• Python 3.10+
• Git installed
• Groq API key

Get your API key here:

https://console.groq.com

---

1. Clone the Repository

git clone https://github.com/adityajhinjha/ai-code-review-agent.git
cd ai-code-review-agent

---

2. Create Virtual Environment

Windows

python -m venv venv
venv\Scripts\activate

macOS / Linux

python -m venv venv
source venv/bin/activate

---

3. Install Dependencies

pip install -r requirements.txt

---

4. Configure Environment Variables

Create a .env file in the project root:

GROQ_API_KEY=your_groq_api_key_here

---

5. Run the Application

streamlit run app.py

Open:

http://localhost:8501

Paste a GitHub repository URL and click:

Analyze →

---

Architecture Diagram

┌──────────────────────────────────────────────────────────────┐
│                    app.py (Streamlit UI)                    │
│   URL Input → Analyze Button → Filters → JSON Export       │
└─────────────────────────────┬────────────────────────────────┘
                              │
                              ▼
┌──────────────────────────────────────────────────────────────┐
│               pipeline.py (Orchestrator)                    │
│     Coordinates ingestion → parsing → review pipeline       │
└──────────────┬───────────────────────┬───────────────────────┘
               │                       │
               ▼                       ▼

┌──────────────────────┐    ┌───────────────────────────────┐
│     ingestion.py     │    │          parser.py           │
│──────────────────────│    │───────────────────────────────│
│ • Validate URL       │    │ • ast.parse()                │
│ • Clone repository   │    │ • Extract semantic chunks    │
│ • Find Python files  │    │ • Syntax-error fallback      │
└──────────────────────┘    └───────────────────────────────┘
                                          │
                                          ▼
┌──────────────────────────────────────────────────────────────┐
│                reviewer.py (LLM Interface)                  │
│──────────────────────────────────────────────────────────────│
│ • Prompt Engineering                                        │
│ • Parallel AI Reviews                                       │
│ • Confidence Scoring                                        │
│ • Repository Summary Generation                             │
└──────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌──────────────────────────────────────────────────────────────┐
│                    Streamlit Dashboard                      │
│──────────────────────────────────────────────────────────────│
│ • Issue Cards                                               │
│ • Severity Filters                                          │
│ • Confidence Filters                                        │
│ • Health Score                                              │
│ • JSON Export                                               │
└──────────────────────────────────────────────────────────────┘

---

File Structure

ai-code-review-agent/
│
├── app.py
│
├── agent/
│   ├── __init__.py
│   ├── ingestion.py
│   ├── parser.py
│   ├── pipeline.py
│   └── reviewer.py
│
├── prompts/
│   ├── review_prompt.txt
│   └── summary_prompt.txt
│
├── requirements.txt
├── README.md
└── .env

---

End-to-End Pipeline

GitHub URL
    │
    ▼
Clone Repository
    │
    ▼
Find Python Files
    │
    ▼
AST Parsing
    │
    ▼
Semantic Chunk Extraction
    │
    ▼
Parallel LLM Reviews
    │
    ▼
Structured JSON Results
    │
    ▼
Repository Health Summary
    │
    ▼
Interactive Dashboard

---

Key Features

AST-Based Semantic Parsing

Uses Python’s built-in ast module to extract:

• functions
• async functions
• classes

instead of sending entire files to the LLM.

---

Parallel AI Reviews

Uses:

ThreadPoolExecutor

to review multiple chunks simultaneously, reducing latency significantly.

---

Confidence-Rated Reviews

Every issue includes:

• severity
• confidence score
• actionable suggestion

Low-confidence findings are isolated under:

Verify This

---

Syntax Error Recovery

Even if AST parsing fails, the system:

• creates fallback chunks,
• preserves exact syntax error locations,
• still allows the AI to analyze broken files.

---

Repository Health Summary

The agent generates:

• overall health score,
• issue breakdown,
• recurring patterns,
• critical issues,
• recommended next steps.

---

Configuration Constants

Constant	File	Default
MAX_CHUNKS	pipeline.py	5
MAX_WORKERS	pipeline.py	5
MIN_LINES	parser.py	5
MODEL_NAME	reviewer.py	llama-3.3-70b-versatile

---

Known Limitations

1. Duplicate chunk reviews — ast.walk() extracts both a class and its methods as separate chunks, so method code gets reviewed twice, wasting API calls and producing conflicting findings.
2. Limited chunk review count (MAX_CHUNKS = 5)
3. No persistent database
4. Small functions silently skipped — Any function under MIN_LINES = 5 is never reviewed with no warning. A one-liner like def get_password(): return "hardcoded123" would be completely missed
5. No retry/backoff system
6. No clone timeout — Repo.clone_from() has no timeout. A large repository will block the UI indefinitely with no way to cancel.
7. No GitHub PR commenting yet

---

Future Improvements

Multi-Language Support

Add support for:

• JavaScript
• TypeScript
• Go
• Rust

using tree-sitter.

---

GitHub Pull Request Integration

Post inline comments directly onto pull requests.

---

Incremental Reviews

Review only changed files/chunks using hashing and caching.

---

Distributed Architecture

Add:

• Celery
• Redis
• background job queues

for multi-user scalability.

---

Analytics Dashboard

Track:

• issue trends,
• repository health over time,
• recurring problematic files.

---

Health Score Reference

Score	Label	Meaning
90–100	Excellent	Minor or no issues
70–89	Good	Few medium issues
50–69	Fair	Multiple medium issues
30–49	Poor	Several high-severity issues
0–29	Critical	Fundamental problems

---

Dependencies

Package	Purpose
streamlit	Frontend UI
openai	Groq OpenAI-compatible client
python-dotenv	Environment variable loading
gitpython	GitHub repository cloning

---

License

This project is open-source and available under the MIT License.

---

Author

Aditya Jhinjha

GitHub: https://github.com/adityajhinjha