Portfolio
Public projects with real repos and real live URLs
I wanted this to read more like a working index than a polished gallery. Everything here links to a public GitHub repo and a live public URL, so you can inspect the code and the product instead of taking my word for it.
Workflow Garden
A public learning site for issue-driven AI development workflows.
Workflow Garden turns a fairly opinionated software workflow into something a newcomer can actually browse. It mixes evergreen articles, project pages, concept pages, and a generated activity diary, so the site feels alive rather than frozen in one explanatory moment.
How it works
The site is built with Next.js and Velite, then fed with generated content from local repo activity. A refresh pipeline creates static JSON and search indices before deploy, which lets Vercel serve the whole thing without runtime access to the local Code workspace.
Proof Pack
A launch-review app that turns a URL into a concise critique pack.
Proof Pack is for that annoying moment right before launch, when a page technically works but still needs someone to say whether the hierarchy is clear, the mobile layout holds up, and the whole thing actually feels ready to ship.
How it works
A user submits a public URL and the app builds a review pack around clarity, accessibility, mobile risk, and concrete next steps. The product is intentionally lightweight, but the thinking behind it is not.
Choice Compass
A weighted decision helper for comparing a small set of options.
Choice Compass is a small app with a very specific job: help someone compare a few real options without sliding straight into spreadsheet sprawl. It is deliberately simple, fast, and browser-local.
How it works
The app lets a user define options, add weighted criteria, score each option, and review the winner with a shareable summary. Everything runs in the browser, with `localStorage` handling persistence, so there is no backend to look after.
Model Intelligence Maintainer
A workbook and static guide for comparing model quality, price, and provider coverage.
This project pulls together model metadata and benchmark signals from multiple sources, then turns them into a workbook and static guide for a very practical question: which model should I actually use?
How it works
A Python pipeline refreshes and normalizes OpenRouter, Artificial Analysis, Vals, and LiveBench data into deterministic datasets, a workbook, and a deployed static site. The repo keeps provenance, cohort rules, and mapping diagnostics out in the open instead of pretending that part is easy.
rajeevg.com
The personal site and publishing system behind this portfolio.
This site is both a public homepage and a small publishing system. It is where I publish writing, explain projects, and turn technical work into something readable for people who were not in the terminal with me.
How it works
The site runs on Next.js with Velite as a typed content layer, shadcn/ui components for reusable UI, and MDX for long-form posts. That makes it easy to mix structured React components with essays, diagrams, and code-heavy write-ups.