DisProject — AI Code Review Pipeline

The aim of this project is to design and justify an architectural framework and prototype pipeline for integrating multiple LLMs into a structured code review workflow for a single programming language, drawing on prior work in modern code review, LLM assisted programming, and software pipeline design.

• Dynamic pipeline builder via a Streamlit web interface
• Multi-agent architecture (logic, security, performance, coordination)
• Hybrid LLM support (local + cloud)
• Integration with static analysis tools
• Interactive review with diff-based suggestion approval
• Batch testing and evaluation tooling

• Python 3
• Ollama (for local LLMs)
• API key (for cloud-based agents)

Clone the repository and install dependencies:

git clone https://github.com/SamHurst47/DisProject.git
cd DisProject
pip install -r requirements.txt

Create a .env file in the root directory:

GEMINI_API_KEY=your_key_here

Install Ollama to run local models such as llama3.

Launch the Streamlit UI to build and run pipelines interactively:

streamlit run app_ui.py

Run these from the root directory:

• optimising_pipeline.py Used for prompt tuning and pipeline configuration experiments

python optimising_pipeline.py

• testing_suite.py Executes batch experiments across multiple pipeline configurations and exports results to CSV

python testing_suite.py

• app_ui.py — Streamlit interface for pipeline configuration and execution
• main.py — Lightweight console-based execution manager
• requirements.txt — Python dependencies

• pipeline.py — Core orchestration engine
• llms_modules.py — LLM agents (logic, security, performance, coordinator)
• static_modules.py — Static analysis wrappers (Pylint, Bandit, regex tools)
• input_modules.py — Input handling (text/file ingestion)
• output_modules.py — Output formatting and delivery

Run the test suite:

pytest test_pipeline.py

The testing_suite.py script was used for large-scale experimental evaluation across pipeline configurations.

By using this system, you agree to the following:

• Do not upload code containing personal or sensitive data
• You must have the legal right to analyse any submitted code

• The system must not be used for exploitative purposes
• This project depends on third-party tools and APIs; ensure compliance with their licenses (e.g., Gemini API usage limits)
• Commercial use will require additional licensing review

• LLM outputs are not guaranteed to be correct or unbiased
• The system is intended to assist, not replace, human code review
• Over-reliance may reduce manual review skills