RoseReview AI

Landing Page

💡 The Inspiration

Our background in web development and enthusiasm for data science led us to take a critical look at modern developer tooling. Having previously built tools that integrate AI APIs, such as RepoRover AI, we noticed a glaring gap in the market: most automated PR bots act like basic linters. They flag missing semicolons but completely miss the larger architectural context.

We wanted to build something significantly better. We envisioned an AI that acts like a senior engineer on your team—someone who is empathetic yet strict, who evaluates architectural boundaries, blast radius, and deployment risks. That vision became RoseReview AI.

⚙️ What it does

RoseReview AI is a repository-aware code review platform that provides deep context analysis rather than surface-level linting.

Instant PR Tracking: Paste any GitHub PR URL into our dashboard to instantly sync the repository data.
The "Senior Dev" Review: It evaluates the Overall Health Score (security, performance, maintainability), checks for dependency coupling, and calculates the deployment risk (Blast Radius).
Seamless Integration: Once the AI generates its strict, humanized markdown review, you can post it directly to the GitHub PR thread with a single click.
Premium Experience: Features a stunning, custom dark-mode dashboard with dynamic context switching and sleek skeleton loaders while the AI runs its analysis.

🛠️ How we built it

To guarantee blistering performance, we architected this as a lightweight monorepo:

Frontend: We went completely zero-framework. The dashboard is pure HTML5, CSS3, and ES6 Modules. We built a custom design system using CSS variables, Flexbox/Grid, and vanilla JavaScript to handle global state management and micro-animations.
Backend: We utilized Fastify and TypeScript for our API layer to keep the server execution lightning-fast.
Integrations: We rely on the GitHub REST API (Octokit) to fetch PR contexts and post reviews, while the Groq API powers our rapid AI inferences. We leverage SQLite and Redis for local state and caching.
Deployment: The entire system is pre-configured for Vercel, which serves our static Vite frontend and compiles our Fastify backend into Serverless Node.js functions.

🚧 Challenges we ran into

Building a highly interactive, dynamic dashboard without relying on heavy frameworks like React was a massive technical hurdle. We had to architect our own global state management system from scratch using vanilla JavaScript.

Additionally, prompt engineering proved surprisingly difficult. It took extensive iteration to force the AI to stop sounding like a generic bot and start delivering the high-signal, strict-but-empathetic tone of a real senior developer. Safely handling GitHub tokens and navigating API rate limits also required careful architectural planning.

🏆 Accomplishments that we're proud of

We are incredibly proud of the performance and aesthetics of our zero-framework frontend. Achieving smooth skeleton loaders, dynamic state updates, and a premium dark-mode UI using just native web technologies is a huge win. Most importantly, we're proud of the AI output—it genuinely feels like getting a thoughtful code review from a human mentor.

📚 What we learned

We learned that stripping away heavy frontend frameworks and returning to vanilla web technologies is sometimes the best way to deeply understand browser capabilities and maximize performance. On the backend, we discovered how powerful and efficient Fastify paired with TypeScript can be for rapidly proxying complex API requests and integrating with Octokit.

🚀 What's next for RoseReview AI

This is just the beginning. We plan to introduce AI-generated code fixes that developers can commit directly from the dashboard. We are also exploring smart test generation to automatically write unit tests based on the PR delta, and an advanced visual map for merge impact analysis so teams can visually track how a pull request will affect their broader ecosystem before they merge.