LLM-Driven Code Review Agent
AI-powered web app that performs structured, explainable code reviews using Google Gemini to improve code quality, clarity, and performance.
React (Vite)Node.jsExpressGemini 2.0 FlashPrompt Engineering
Problem
- Manual code reviews can be time-consuming, inconsistent, and dependent on reviewer availability.
- Developers often don’t get clear explanations for why changes are needed, slowing learning and productivity.
Solution
- A web-based service that accepts code, sends it to Gemini with a strict prompt, and returns sectioned review feedback.
- Focuses on explainability and practical developer workflows (UI-friendly structured output).
Key Features
- Automated code analysis using LLMs
- Predictable sectioned output (Issues, Improvements, Optimizations, Explanation)
- Multi-language support (Python, JavaScript, Java) and extensible design
- Clean separation of frontend and backend; API-first architecture
- Designed for future CI/CD + GitHub PR integration
Screenshots
Tip: Click any image to preview.
Technical Deep Dive
Biggest challenge
Designing prompts that consistently produce structured, actionable code review feedback while avoiding hallucinated or unsafe suggestions across languages.
Structured output
Enforced fixed headings and explicit constraints in the prompt so the UI always renders predictable sections.
Next Improvements
- GitHub Pull Request integration for automated reviews on PRs
- Scoring system for security / performance / readability