Full-stack AI engineer
I take AI work from the model layer through the interface and out to a live URL: retrieval, agents, evaluation and local inference, wired into business systems that already exist — then the front end on top and the deployment that ships it. Thirty years building production systems; the last two spent putting language models inside them.
The skill set
Every number on this page is measured off the running system, not estimated. Where a figure couldn’t be verified against something on disk, it isn’t here.
Selected work
Indexes every past session into Postgres, scores them with a local LLM judge, embeds the keepers, and injects the relevant ones back into new sessions automatically. Retrieval is vector search fused with full-text by Reciprocal Rank Fusion. A separate audit layer extracts testable claims from past answers, checks them against the filesystem, and flags the false ones on recall — the system catches its own hallucinations.
A local model works a backlog across parallel project lanes overnight while a frontier orchestrator reviews and promotes the work. Orchestrator and workers are strictly separated — workers verify but never write. A verdict only counts if it carries an exit code, a quoted log tail and a file reference, so unproven passes get demoted. Includes atomic task claiming with stale-claim recovery and three-strike escalation to a human.
A production platform where every AI task defaults to a local model and only routes to a frontier provider when a tenant admin explicitly opts that task in. Before anything can leave the box, a local NER model running in an isolated subprocess finds and substitutes personal data, which is rehydrated on the way back. Natural-language reporting compiles to SQL behind a safety guard, with a stored-query fast path.
A drop-in replacement for OpenAI in a production deployment: the legacy platform points at it unchanged. Four task-specific skills are routed by the requested model name, with no silent fallback on an unknown one. The SQL skill is grounded with few-shot examples retrieved by cosine similarity from an embedded corpus, and degrades gracefully if the vector store is missing. Zero third-party dependencies.
Unattended bakeoffs that gate adoption on the things that actually break an agent loop. A candidate 35B model was disqualified outright for failing tool-calling before it ever reached production; survivors were throughput-benchmarked on identical hardware. Results include an honest limitations section and a documented negative result rather than a buried one.
Generates natural-language→SQL training pairs by wrapping the live production system prompt and schema around a local 35B model, so every pair is in the exact dialect the real system expects. Generated SQL is validated against an in-memory database before it’s kept, and prompt changes are scored as pass-rate against a frozen validation set.
The front end
Cinematic scroll-worlds on a video engine I wrote, real-time 3D in Three.js, and games — all built, deployed, and responding right now. Every title below is a link.
Cinematic scroll-worlds — pre-rendered camera flights, scrubbed by scroll
This CV, as a worldskills.sitotis.ai
Thirty years, one takeCareer fly-through
Colosseum, 107 ADHistorical world
PompeiiHistorical world
TriumphCanvas 2D + video
siTOTisCompany front door
WorldsHub
Papercraft dioramaClient concept
Isometric dioramaClient concept
Real-time 3D — Three.js / WebGL
The Quiet WingExplorable 3D gallery
The WoodsProcedural 3D forest walk
Museum of PlayWalkable games museum
The ArcadeWalkable games portal
Games — Canvas 2D, one mechanic each
DRIFTTwo buttons
TETHERTwo buttons
FATHOMOne button
VIGILChoose what to lose
COUNTERPOINTSoundtrack is the score
The Card RoomNext.js party games
Product & editorial
Experience
Design and ship applied-AI systems end to end: local LLM inference on Apple Silicon behind an OpenAI-compatible gateway, retrieval-augmented systems on Postgres and pgvector, multi-agent orchestration, evaluation harnesses, and natural-language reporting compiled to SQL — integrated directly into the business platforms clients already run on.
Migrated an aviation media publisher off shared spreadsheets onto a custom platform: quoting and invoicing, a WebDirect client portal for order confirmation, and a multi-publication ad-placement module. Consolidated five contact databases — over 130,000 records — into one system, and replaced their accounting package.
Developer at one of the largest FileMaker-only premium partners in the world. Built and maintained solutions for clients including two Fortune 500 companies; one of three developers building to a Material Design standard.
Built cloud-hosted CRM systems from concept to production, including HTML email delivery from within the application, WebDirect access and custom reporting. Also responsible for server hardware and software, daily cloud backups and SSL maintenance.
Designed, integrated and automated the school's data systems with stakeholders across the organisation — system and network modelling, planning and implementation — and continued development, administration and training for its custom Student Information System.
Built a custom Student Information System from concept to production with the VP of Curriculum: admissions, attendance, reporting, family demographics, parent–teacher interview booking, course evaluation, auto-enrolment and OUAC/OnSIS integration. Delivered training in person, online and by video.
Taught JavaScript, form validation, server-side scripting in PHP and introductory SQL database design. Final project: a student-built PHP/MySQL content management system.
Head administrator for a board-wide web-based student information system: deployment, administration and Ministry of Education reporting via XML customisation. Trained principals, administrative staff, teachers and students. Also built an issue-tracker application with mobile access for on-site technicians.
Built e-commerce systems on Propel's internet infrastructure platform — owned all search components and the order-management area — and helped convert the application from a JSP/Servlet architecture to the Struts/Template framework.
Co-developed a subscription internet platform offering e-commerce, message board and trading-post modules, and built an issue-tracking application for Nortel Networks.
Supervised lab technicians and ran a mixed Macintosh and Windows NT 4.0 network for students and faculty.
Diagnosed and repaired client hardware and software, and staffed a paid technical support line.
Tools
Education & certification
Open to
Forward-deployed and solutions engineering, applied AI, or AI-heavy product work. Based in Ontario, Canada. Everything above was built remotely, alone, from this desk.
AUTO_HERMESAutonomous coding loop
Five Hours BackAI assessment offer
Compliance auditData analysis & deck
FridgeMicro-site