
Torrin Leonard
Product Engineer · Available nowI build and lead production AI software 0 to 1, end to end.
Open to remote roles · Located in Waterloo 🇨🇦
Experience
Full Stack Software Developer
Mar 2023 to Sep 2026Tech: Node.js/Express, React, Next.js, TypeScript, MongoDB, Python/FastAPI, OpenAI, Qdrant, ChromaDB, GeoPandas, Docker, GitLab CI/CD, AWS, Linux- Built a 0 to 1 AI proposal product as its sole engineer through the initial build, then led it as the team grew to seven, authoring the majority of commits and 64% of the code.
- Rebuilt the marketing site on Next.js and owned it end to end through CI/CD and release, growing organic search from under 1,000 clicks and under 1M impressions to 52.1K clicks and 6.17M impressions in six months.
- Authored and merged 70 production releases on the AI product, and served as release manager for a main-product release, coordinating the rollout through pre- and post-push testing.
- Built the RAG ingestion and retrieval pipeline on OpenAI embeddings and Qdrant, migrated off ChromaDB, using query-expansion retrieval rather than single-vector search.
- Implemented a client model fine-tuning pipeline on the OpenAI fine-tuning API, and a proxy metering token cost per user on every model call.
- Built an admin prompt orchestration layer so non-engineers tune production AI behaviour live, with versioned code-defined fallbacks and prompt-injection and hallucination guards.
- Integrated server-hosted GIS segmentation models (SamGeo, LangSAM) for satellite imagery selection in the main app.
- Added Stripe payment systems and supported site-based pricing rollout with product.
- Onboarded and mentored 2 full-time developers, reviewed 55 merge requests, and set written acceptance criteria on work handed off to them.
- Created and decomposed 2 company epics, sat on interview panels, and held the architecture decision on the AI product and the frontend.
- Wrote and maintained the team developer handbook, commit and merge-request policy, and engineering principles with a self-review checklist.
Co-Founder & CEO, Lead Software Developer
This Cozy Studio
- Founded and ran a federally incorporated 3D modeling and rendering contract pipeline studio across 7 client engagements with no outside funding; owned product, clients, contracts, and delivery end to end.
- Built per-client combinatorial rule engines and unattended batch rendering on rented GPU capacity, delivering over 50,000 rendered files to one client in under a week.
- Open-sourced the studio's Blender tooling (see Projects) and grew a companion tutorial channel to 282,388 views and 3.29K subscribers across 25 videos, driving adoption.
Automation & Accessibility Engineer (Contract / Part-time)
Mar 2021 to May 2022Tech: JavaScript/TypeScript, TestCafe, Node/npm, Git, Jira, WCAG- Built TestCafe regression suites in TypeScript/JavaScript, reducing manual cycles and improving release confidence.
- Performed WCAG accessibility audits and partnered with engineers/PMs to ship compliant releases; triaged defects within CI workflows.
Automation & Accessibility Engineer (Contract / Part-time)
Jun 2013 to Mar 2021- Delivered manual and automated testing as a contractor for clients including TunnelBear, Hubba, THRILLWORKS, and Hopscotch, building reusable test tooling and cross-browser checklists for WCAG-aligned releases.
Projects
Public directory of Waterloo-region tech startups (formerly KWBuilds), with news, events, jobs, a forum, and a public MCP server.
- Built solo across 483 commits: a live directory of 824 published startups scraped and enriched from Communitech, Velocity, and other regional hubs, with news, events, and jobs subsystems running on the same pipeline machinery.
- Chose browserless scraping (got, cheerio) after a measured bake-off across 493 sites recovered content for 98.8%, retiring the headless browser and Docker from the stack.
- Serves a public MCP server with a hand-rolled OAuth 2.1 authorization server (Dynamic Client Registration, PKCE) so claude.ai and ChatGPT connectors can attach while anonymous clients still work.
No-code web platform for behavioral research: researchers build an experiment visually, share one link, and get charts and raw CSV back.
- Compiles each experiment to a constrained JSON definition, validates and freezes it server-side against a generated op whitelist, then runs it in participants' browsers on a purpose-built interpreter rather than generated code.
- Runs author-written code in an opaque-origin sandboxed iframe with no network access, and measures reaction times on a monotonic high-resolution clock so wall-clock changes cannot corrupt them.
Sandboxed execution platform for agent-authored modules, running live.
- Agent-authored modules stored as Postgres rows, each run in an isolated child process that cannot reach the filesystem or its peers.
- An egress proxy injects credentials outside module code, so agent-written modules never hold a plaintext secret.
Full-stack TypeScript toolkit for apps whose state is an observable document, published to npm under MIT as @aweftjs.
- Built solo across 399 commits: 23 packages covering state, sync, storage, server, UI, sandboxing, jobs, and auth. Builderloo, maccobylab.com, mcp.torrin.me, and torrin.me all run on it.
Blender add-on automating procedural 3D asset generation and export, written against the Python/Blender API.
- 1,015 stars and 324 forks, used by multiple studios. Authored 423 of 496 commits across 7 releases.
Reactive UI component library, published to npm.
- Lead developer of a 70 component library published to npm under MIT, 619 of 1,105 commits. It powers Equator's mapping and AI platform and OpenGig.org.
Open-source platform for gig workers and customers.
- Built solo across 581 commits: full-stack JavaScript with a custom UI framework, database state-sync, and a WebSocket layer, deployed on DigitalOcean behind NGINX with GitHub Actions CI.
Blender add-on bringing Stable Diffusion texture generation into Blender.
- 116 stars, one of five open-source Blender tools totalling 1,168. Shipped Sept 2022, two weeks after Stable Diffusion's public release and three months before ChatGPT, with the studio hosting the weights.
Local music player with AI-assisted lyrics and metadata.
- Uses Whisper to auto-generate and align lyrics for time-synced playback.
- Displays time-synced lyrics in an Apple-style karaoke mode lyric scroller.
- Enhances albums with additional metadata like artwork and descriptions.
Skills
- Languages: JavaScript, TypeScript, Python, Rust
- Frontend: React, Next.js, destamatic-ui, HTML/CSS
- Backend: Node.js, Express, FastAPI, Django, REST APIs, WebSockets
- Databases: MongoDB, Qdrant, ChromaDB
- AI & Data: OpenAI API (embeddings + fine-tuning), RAG (Retrieval-Augmented Generation), vector search, evals, prompt injection defense, Whisper, pandas
- AI-assisted development: Claude Code and OpenCode daily, plan first, every diff reviewed and tested before it ships
- Cloud & DevOps: Docker, GitLab CI/CD (Continuous Integration/Continuous Deployment), GitHub Actions, DigitalOcean, AWS (Amazon Web Services), Proxmox, NGINX, Linux (Ubuntu/Arch)
- Practices: Release management, code review, mentoring, technical interviewing, WCAG accessibility audits
- Testing: TestCafe, automation and regression testing
Education
Self-taught software engineer.
- CS50x, Introduction to Computer Science (Harvard University, 2025)
- Introduction to Web Accessibility, WCAG (W3Cx, 2021)
- ANU-ASTRO1x, Greatest Unsolved Mysteries of the Universe (Australian National University, 2020)
- 16.00x, Introduction to Aerospace Engineering (MIT, 2021)
Interested? Let's talk!
Fill out the form below, email me directly, or dm me on LinkedIn. Either way I'll get back to you quickly!

