Raphael Samuel
Building AI agent tooling and data infrastructure end to end. Two products of my own, plus client work.
About
A quick intro.
Six years of writing Python, Go, TypeScript, and Next.js, most of it spent on two questions: does this hold up once real people depend on it, and who fixes it at 2am when it doesn't. Quorel and VarsityLine are both still answering yes.
Selected Work
Projects
Python
Go
TypeScript
Next.js
Tailwind
MySQL
PostgreSQL
GoNext.jsTypeScriptMySQLQuorel
Gives your AI agent a live, versioned view of any website. Describe the fields you want in plain English, Quorel extracts them nightly and keeps every past version permanently queryable. Every dataset ships with an MCP server, so Claude or any MCP-compatible agent can query it, catch what changed since last week, and even fix bad rows itself, publishing the correction as a new, reversible version.
Next.jsTypeScriptPostgreSQLTailwindVarsityLine
Nigerian students figure out admission cut-offs from group chats and outdated PDFs. VarsityLine fixes that: verified cut-off marks, courses, and screening dates for every Nigerian university, cross-checked against each school's own admissions office, with a "last confirmed" timestamp on every entry. Search by university, by course, or both with state and type filters, export results as a PDF, and subscribe for access to search and compare feature for decision making. Ships with a custom MCP server so Claude Desktop can autonomously monitor the source dataset on Quorel, clean changes, and push validated updates across all 24 universities.
Go
SQLiteMySQLPostgreSQLGnat
Existing self-hosted analytics tools force a tradeoff between depth and simplicity. Gnat is a single Go binary, no Docker, no multi-service stack, with funnels, cohorts, retention curves, and auto-discovered path analysis built in. Every instance ships with a native MCP server, so AI agents can query analytics data directly, no separate export step required.
Python
JavaScriptSkim
Uploading a 40-page PDF and getting a wall of text back isn't a summary, it's a chore. Skim lets you tell it exactly what you're looking for, then returns a clear, page-by-page summary in seconds, built for students and teachers who need the signal, not the whole document.
TypeScriptNext.jsWildpalace SEO PLP
Store owners waste hours hand-writing SEO copy for every product listing page, and most of it still doesn't rank. Wildpalace plugs into Shopify and auto-generates SEO-optimized Product Listing Pages straight from keyword intent, turning a manual content chore into something that runs itself.
More experimentation and older work on my GitHub if you want to dig deeper.
See more on GitHub ↗Frontend & UI
Design Work
Featured: 3D Viewer
Interactive 3D Ring Viewer
A real-time 3D ring viewer for jewelry e-commerce. Customers rotate, zoom and inspect rings from every angle before buying, reducing returns and building confidence.
Next.js
Three.jsTypeScriptTailwind
WebGLLanding Pages
Jewelry Store Collections
Next.jsTypeScriptTailwind
Framer MotionAqua Collection
Cool-toned jewelry store with a crisp aqua palette. Showcases rings and accessories in an airy, modern layout with smooth hover transitions and a streamlined checkout experience.
Why Work With Me
Not just another developer.
Recent Writing
From the Blog
A storage problem, a clustering fix, and the race condition that fix caused. How one bad graph led to a better one.
2026-02-12After two projects with REST endpoints that nobody documented, I gave tRPC a shot. Here's what changed.