Collins Thomas

RampLab reads a repository and writes the book it never had: an interactive edition of chapters, figures, and checkpoints that takes a new engineer from first clone to first contribution.

Every sentence cites the file and lines it came from, and every citation is checked mechanically before an edition ships. If the code cannot back a claim, the claim does not print.

Editions answer questions from their own pages, with receipts. You can press your own on the Claude Code login you already have, and your code never leaves your machine.

• AI
• Claude Code
• Developer onboarding

Visit ramplab.dev Read an edition

A launch campaign for astragon Entertainment and Simteract. Players sign in with Steam and earn points by watching trailers, wishlisting, following, playing the demo, taking a quiz, referring friends, and pre-ordering on Steam, Epic, PlayStation, or Xbox.

Every point lands in a ledger the player can read. Rewards unlock automatically at each threshold, from a wallpaper to a DLC key to a full game key. Built for Achievo, where every quest feeds the wishlist and playtime signals Steam's algorithm rewards.

• Full-stack
• Steam
• Gamification
• English, German

Visit the campaign

A gamescom 2026 campaign for four 4Divinity titles: The Defiant, Showa American Story, Guns of Eschaton, and Genigods: Nezha. Quests across all four games pay into one XP balance.

It mixes online quests with booth quests in Cologne. Players scan QR codes at the stand to unlock the top tiers, so visitors and players at home share one progression, with rewards up to an XPG hardware ballot and game keys.

• Full-stack
• Steam
• Events
• English, German

Visit the campaign

A pre-launch campaign for a survivors-like mecha shooter coming to PC in 2027. Six quests span wishlisting on Steam, joining Discord, watching the trailer, a quiz, referrals, and playing the demo at the gamescom booth.

Points unlock four reward tiers: wallpapers, the soundtrack and 3D print files, a Steam key draw, and a mystery gift collected at the booth.

• Full-stack
• Steam
• Discord
• English, German

Visit the campaign

Activity Monitor shows seventeen anonymous processes. memhogs walks the process tree so every helper, renderer, and language server rolls up into the app that spawned it: one app, one row, one total.

It reports memory footprint on macOS and PSS on Linux, so shared memory is never counted twice. It resolves real app names, stops at interactive shells so your own jobs stay yours, and ships as one 2.2 MB static binary with JSON output and a live watch mode.

brew install cicerothoma/tap/memhogs

• Go
• macOS
• Linux
• Open source

Try the live demo View on GitHub

Every app on the phone, ranked by honest memory, with helper processes grouped under the app that owns them. Tap an app to see its processes.

A safe reclaim button kills background processes only, the same thing Android does under memory pressure, and reports exactly how much memory came back. Nothing you have open is touched.

Android hides other apps' memory from normal apps, so memhogs uses Shizuku to read it the way a USB debugger would. No root needed. Requires Android 8 or newer.

• Kotlin
• Android
• Shizuku
• Open source

Download the APK View on GitHub

A web tool that migrates music libraries across the four major streaming platforms, with no signup. It matches tracks across catalogues automatically and lets you preview every match before you transfer.

• Web
• Music APIs

Visit streamswap.app

Create reusable, customisable forms and share them with one link to collect or share verified information: applications, rentals, onboarding. Built around clean data, exports, and a smooth multi-party workflow.

• Backend
• Forms
• Data

Visit getgopath.com

The Node.js backend for a peer-to-peer FX marketplace: escrow, bank and wallet management, role-based admin access, and detailed activity logging. A modular monolith with real-time updates.

• Node.js
• Escrow
• RBAC

Open Kaorex

The backend for loans, repayments, payments, and investment features used by tens of thousands of people. Built for reliability, risk-aware flows, and clean integration with payment and notification services.

• Backend
• Payments
• Lending

Visit mylendme.co

A computer vision model trained with transfer learning on EfficientNetB4 to classify Nigerian dishes into 18 classes, deployed as an interactive demo. I prepared the dataset, trained and evaluated the model, and documented where it struggles, such as telling similar soups apart.

• Python
• TensorFlow
• Computer vision

Try the demo

A Slackbot that uses retrieval-augmented generation over internal documents to answer HR and policy questions. It handles document ingestion, retrieval, and response generation, all inside Slack.

• RAG
• LLMs
• Slack

Open the dashboard