I'm a software engineer in Munich. I build native apps, open-source tools for coding agents, and scientific software.

Before that I did a PhD in materials science and spent two years at Diamond Light Source, the UK's national synchrotron, building instrumentation and deep-learning tools for X-ray imaging.

tristanmanchester.com · LinkedIn · tristan@tristanmanchester.com

Differentiable tomography in JAX for parallel-beam, laminography and lab cone-beam CT. It does FBP/FDK and iterative reconstruction, solves the volume and every view's pose together, and has a CLI that takes NeXus, APS, Nikon or TIFF scans through to slices. Above: a simulated watch movement, and the same scan with a wobbling sample before and after alignment. Docs at tomojax.dev. Probably the most technically interesting thing I've built.

Built with: Python, JAX, Pallas, CUDA, HDF5

Fast local code search for AI agents. Point it at a repo and ask "where is authentication handled?" and it returns ranked files, line ranges and code chunks, using BM25, semantic or hybrid ranking. Warm queries take 2.7 ms. Runs as a CLI, a Rust crate or an MCP server, and installs from crates.io or Homebrew.

Built with: Rust, Model2Vec, Tree-sitter

A private clipboard memory for macOS. It saves everything you copy to a local SQLite archive and lets you search, recall and restore it from the terminal, the menu bar, or an agent like OpenClaw. Copied screenshots can be made searchable with on-device OCR. On crates.io and Homebrew.

Built with: Rust, SQLite, objc2, Apple Vision

You've got a laptop, a pile of scrap, and zero internet. Can an LLM help you survive without getting anyone hurt? 500 survival questions, judge-scored rubrics, an offline Wikipedia retrieval track, and a dashboard for exploring runs. Full writeup here.

Built with: TypeScript, Next.js, Rust

An animated dot-matrix display for React Native, inspired by the ElevenLabs UI. Skia rendering, 60fps UI-thread animations with Reanimated, a VU meter mode for live audio, and a bunch of built-in presets. On npm.

Built with: React Native, Skia, Reanimated

• rheo: instant macOS Space switching. Swipe or use a shortcut and the Space changes without the slide animation. Rust core, native AppKit, and a CLI.
• agent-skills: 36 skills for Claude Code, Codex and other agents, from App Store readiness audits to Skia animation.

My most successful (and simplest) app, a notes app built around Andrej Karpathy's note-taking technique. Everything goes into one stream, older notes sink, and a swipe rescues whatever still matters. People seem to particularly like that there's no subscription, and no AI. Runs on iPhone, iPad, Mac and Android, with a CLI for agents.

App Store · Mac App Store · Google Play
Built with: SwiftUI, React Native Expo, TypeScript, SQLite, Rust

A tiny shared to-do list for two. Couples, flatmates and families get one list with assignees, repeating chores and home-screen widgets, and nothing else to manage.

App Store · Google Play
Built with: React Native Expo, Convex

A weekly check-in app for couples. Jot things down during the week, talk them through together in one guided check-in, and leave with a shared recap and a small pact for the week ahead.

App Store
Built with: React Native Expo, Convex, Skia

Paste a recipe, drop a link, photograph a cookbook page or just read it out loud, and RecipeCrow adapts it to your diet, what's in your cupboard and your calorie target. Every dish gets a pixel-art portrait, and the kitchen outside the window follows your local time and weather.

Built with: Next.js, Convex, Clerk

My least successful (and most complex) app, an AI voice-notes app which is the antithesis of Gravity. Talk through what's on your mind and get back a structured note with headings, a transcript and your to-dos pulled out. People seem to particularly like not using it.

App Store · Google Play
Built with: React Native Expo, Convex, Clerk, Vercel AI Gateway, RevenueCat

At Diamond I built segmentation pipelines for synchrotron X-ray tomography. The reconstruction work later became TomoJAX.

ParticleSegmentation	ScrambledSeg
k-means and watershed segmentation	SegFormer deep-learning segmentation

Based in Munich, Germany 🇩🇪