AI Repository-Aware Code Reviewer

An AI-powered repository-aware coding assistant built using Streamlit, ChromaDB, Sentence Transformers, AST parsing, and Google Gemini AI.

Upload Python files or complete repository ZIPs to perform AI-powered code reviews with semantic retrieval, repository-aware reasoning, static analysis, embeddings, and Retrieval-Augmented Generation (RAG).

The system intelligently analyzes code structure, indexes repositories into vector memory, retrieves semantically related functions across files, and generates contextual AI reviews.

---

Features

• AI-Powered Code Reviews
• Multi-File Code Understanding
• Repository ZIP Ingestion
• Recursive Repository Traversal
• AST-Based Code Parsing
• Static Code Analysis
• Semantic Code Chunking
• ChromaDB Vector Memory
• Code Embedding Generation
• Repository-Aware Retrieval
• Cross-File Semantic Reasoning
• Streaming AI Responses
• Real-Time Repository Indexing
• Semantic Similarity Search
• Metadata-Enriched Retrieval
• Local Embedding Inference
• Modular AI Pipeline Architecture

---

Tech Stack

Frontend

• Streamlit

AI / Retrieval Components

• Google Gemini API
• Sentence Transformers
• ChromaDB
• Transformers
• ONNX Runtime

Code Intelligence

• Python AST
• Static Analysis
• Semantic Chunking

Backend / Infrastructure

• Python
• Recursive File Traversal
• ZIP Repository Extraction
• Temporary Workspace Handling

---

Project Architecture

Repository ZIP Upload
        ↓
Repository Extraction
        ↓
Recursive File Traversal
        ↓
Python File Discovery
        ↓
AST Parsing
        ↓
Semantic Chunking
        ↓
Embedding Generation
        ↓
ChromaDB Vector Storage
        ↓
User Review Request
        ↓
Semantic Repository Retrieval
        ↓
Context Injection
        ↓
Gemini AI Review Generation

---

Repository-Aware RAG Workflow

Repository Upload
        ↓
Code Parsing
        ↓
Function Extraction
        ↓
Semantic Embeddings
        ↓
Repository Vector Memory
        ↓
User Code Review Query
        ↓
Cross-File Retrieval
        ↓
Repository Context Injection
        ↓
AI-Powered Review

---

Folder Structure

CODE-REVIEWER
│
├── embeddings
│   └── chroma_manager.py
│
├── prompts
│   └── review_prompt.txt
│
├── retrieval
│   └── retriever.py
│
├── services
│   └── llm_service.py
│
├── utils
│   ├── ast_analyzer.py
│   ├── code_chunker.py
│   ├── file_handler.py
│   ├── language_detector.py
│   ├── prompt_builder.py
│   ├── repo_ingestor.py
│   └── repo_zip_handler.py
│
├── app.py
├── requirements.txt
└── README.md

---

Installation

1. Clone Repository

git clone https://github.com/decoded15/code-reviewer.git

cd code-reviewer

---

2. Create Virtual Environment

python -m venv venv

Activate Environment

Windows

venv\Scripts\activate

Mac/Linux

source venv/bin/activate

---

3. Install Dependencies

pip install -r requirements.txt

---

4. Add Gemini API Key

Create a .env file in the root directory:

GOOGLE_API_KEY=your_api_key_here

---

Running The Project

Start the Streamlit app:

streamlit run app.py

Application runs on:

http://localhost:8501

---

Static Analysis Features

The system performs AST-powered static analysis for:

• Syntax Validation
• Function Extraction
• Import Extraction
• Long Function Detection
• Deep Nesting Detection
• Import Complexity Analysis

---

Semantic Retrieval System

The application builds repository-aware semantic memory using embeddings and ChromaDB.

Capabilities include:

• Cross-File Retrieval
• Semantic Function Matching
• Repository Context Injection
• Metadata-Aware Retrieval
• Similarity-Based Code Understanding

---

Example Retrieval Flow

auth.py
    ↓
validate_token()

login.py
    ↓
login()

        ↓

Semantic Retrieval

        ↓

AI understands authentication flow
across multiple files

---

AI Engineering Concepts Learned

• Repository-Aware RAG
• Semantic Code Search
• Vector Databases
• Embeddings for Source Code
• AST Parsing
• Static Analysis
• Semantic Chunking
• Cross-File Retrieval
• Context Injection
• Retrieval-Oriented AI Systems
• Repository Indexing Pipelines
• Streaming AI Responses
• Local Embedding Inference
• ONNX Runtime
• Repository Traversal
• Metadata-Enriched Retrieval
• AI Orchestration Pipelines
• Retrieval-Centric AI Architecture
• Local AI Infrastructure

---

Future Improvements

• GitHub Repository Cloning
• Multi-Language Support
• Symbol-Aware Retrieval
• Dependency Graph Analysis
• Repo Maps
• Agentic Debugging
• Local Code LLM Integration
• Ollama Integration
• Autonomous Code Editing
• Tool-Calling Agents
• Terminal Execution Sandboxing
• Memory Systems
• Multi-Agent Coding Workflows
• Docker Deployment

---

Author

Built by Dibyansh (decoded15)