AI Code Reviewer

Lightweight AI code review application powered by Cloudflare Workers AI

---

This project was implemented for the Software Engineering Internship assignment, built with Codex using GPT-5.5 (Medium). The prompts used can be found in PROMPTS.md.

Features

• Structured AI code review with actionable feedback
• Covers multiple aspects of code quality (correctness, security, maintainability, performance, style, documentation, and other)
• Each issue is annotated with an explanation, suggestion, severity, confidence level, and source code position
• Overall rating and summary of the review
• Persistent review storage
• Follow-up chat for follow-up questions and iterative review

Workflow

1. User submits code
2. Backend sends code and system prompt to the LLM
3. LLM returns structured review in a JSON schema
4. Review is stored and displayed
5. User can use the chat to ask follow-up questions or provide additional context

Example of the structured review feedback returned by the LLM:

{
  "summary": "...",
  "rating": 7.5,
  "issues": [
    {
      "type": "correctness",
      "severity": "high",
      "message": "...",
      "suggestion": "...",
      "confidence": "high",
      "location": { "line": 12, "column": 5 }
    }
  ]
}

Architecture

• Frontend (React, Vite, Worker Static Assets): user interface to paste code, view review results, manage sessions, and chat about a review
• Backend (Cloudflare Worker): handles API requests, validates payloads, calls the LLM, and stores/retrieves sessions
• LLM (Workers AI, Llama 3.3): performs code analysis and generates structured feedback
• Database (Cloudflare KV): stores reviews
• Client Storage (Local Storage): stores session id (for indexing in KV) and user preferences

flowchart LR
  Start(( )) --> Frontend
  Frontend --> CW["Cloudflare Worker"]
  CW --> WAI["Workers AI"]
  WAI --> Llama["Llama 3.3"]
  CW --> KV["Cloudflare KV"]
  Frontend --> LS["Local Storage"]

Loading

Getting Started

Deployed App

This app is deployed on Cloudflare Workers and can be accessed at:

https://ai-code-reviewer.r1c4rdco5t4-f9d.workers.dev

Local Setup

Requirements:

• Node.js
• npm
• A Cloudflare account

Clone and install dependencies:

git clone https://github.com/rcosta358/cf_ai_code_reviewer
cd cf_ai_code_reviewer/app
npm install

You can set the VITE_USE_MOCK_REVIEW variable in .env.development to true to use a mock review response without calling Workers AI, which is useful for development without consuming AI credits.

Then, login to your Cloudflare account and create a new Cloudflare KV namespace:

npx wrangler login
npx wrangler kv namespace create SESSIONS_KV

Then, update wrangler.jsonc with the generated KV namespace id.

Finally, run the development server and open the app in your browser at http://localhost:5173:

npm run dev

To deploy the app to Cloudflare Workers, run:

npm run deploy

References

• https://agents.cloudflare.com
• https://workers.cloudflare.com
• https://developers.cloudflare.com/workers-ai
• https://developers.cloudflare.com/kv