An R&D lab · Ireland · est. 2026
I build with a fleet of agents.
SID-Labs is my R&D lab. I'm Glen, an architect, and I research, design, build, test and ship real things with a fleet of AI agents. So far that's a finance app, three games, a documentary series, an app for educational settings and a book. Each one is below, with how it was made.
The method
One builder. A fleet of agents.
I set the direction and the agents do the heavy lifting. A neural data-mapping solution I developed links every project, file and decision, hands out autonomous tasks to the right agents, and keeps one memory that every agent works from. No session starts from zero.
-
The builder
Direction
I write the brief, play every build and decide what ships. My name goes on the result.
-
Neural data mapping
The map
A data-mapping layer I developed connects every project, file, decision and lesson, so an agent can find what it needs without being told.
-
Autonomy
Task distribution
Work is split into autonomous tasks and given to the agents best suited to each one. They run in parallel and hand off cleanly.
-
Shared memory
Memory
Handoffs, decisions and every gotcha are kept in one memory that all the agents read and write, so nothing learned is lost between sessions.
-
Cloud-hosted
Frontier models
Cloud-hosted models such as Grok, Claude and ChatGPT handle reasoning, code, writing and review.
-
Localised
Local models
Localised AI models handle generation across text, image, video, audio, music and 3D.
- 01
Brief
A one-page brief and a definition of done, written into the shared memory.
- 02
Distribute
The work is split into autonomous tasks and handed out to agents working in parallel.
- 03
Cross-review
Every build is critiqued by a different model than the one that wrote it.
- 04
Prove it
I only count something as verified when it has been run and seen to work. Reading the code doesn't count.
- 05
Ship
I play it, sign it off and put my name on it.
- 06
Remember
Every gotcha goes into the shared memory, so no agent hits it twice.
Out of the lab
The work.
01 · Live product · Personal finance
Everything you've ever spent, mapped.
Lithifyte turns your bank statements into a live map of people, accounts, categories and shops. It also does budgets, forecasts, debt payoff plans and net worth. All of it runs in one HTML file in your browser, so your statements never leave the tab.
- 0finance servers holding your ledger
- 1readable HTML file, open source (AGPL-3.0)
- 450+self-tests that run every time the page loads
Drag to turn
A forward-march battler. You drag a growing host of archers across a wide road. Weapon crates wait on one side and recruit gates on the other, while an endless horde with champions and bosses that change form comes down the middle.
- 21levels, ending with the Final Stand
- 2players in online co-op, phone to phone
- 10heroes in 3D, each with its own skill tree
The heroes on the left are the game's real 3D models, rebuilt from their painted portraits by a local image-to-3D model.
03 · Co-op roguelite · In development
The Last
Threshold
Four wayfarers. One open region. Nothing comes back the same.
A co-op roguelite for up to four players, built in Godot 4.7. The first region, the Whispering Vale, is built: zoned woods, terraced settlements, three rock escarpments and a river gorge crossed by four bridges.
- 221designed pieces placed in the Vale
- 100+frames a second in a four-player fight
- 4agents working in one repository
Drag to turn
04 · Piano-learning game · Family playtest
Arc Barrian Island
A brick-built island where the controller is a real piano. Plug in a MIDI keyboard and each zone of the island teaches the next step of the piano. Stars and bricks build dino friends, who move into a free-roam home island called Wonderland.
“Find E — the yellow key!”
“Ahoy! Left hand to the low notes!”
Try a key.
Click the keys or use A–K on your keyboard. The home row is mapped the same way as in the game.
- 104levels across 10 zones
- 101songs, each with a band part
- 965narrator lines, voiced by a local AI model
- 7,123automated tests passing
05 · Documentary series · YouTube
The Bible You Missed
The End Times is an eleven-episode documentary series, written, cast, filmed, voiced and scored with AI and mastered in 1080p. The channel has over 200 YouTube Shorts generated alongside it, and a second season, The Watchers, is written in ten episodes.
- 200+YouTube Shorts generated
- 11documentary episodes
- 10more in script
06 · Education app · In development
Happylog
A developing application for educational facilities.
Happylog is a tablet-first daily logbook for early-years and educational settings. Educators record attendance, meals, sleep checks, toileting, medicine, incidents and milestones on one shared tablet. Every entry is signed with that educator's own PIN. It works offline first and keeps the records an inspection asks for.
- Attendance
- Meals
- Sleep checks
- Toileting
- Medicine
- Incidents
- Milestones
- Visitors
- 1shared tablet per room
- 1PIN per adult, on every entry
- 15minute sleep-check timer that turns amber when a check is due
07 · Book · SID-Labs Press
AI The New Normal
A Practical Guide to Using AI for Research and Knowledge Work
Glen Jordan · Second Edition · 2026
AI accelerates processing.
The Claim Rule, the whole book on one line
Humans own claims.
A field manual for researchers and knowledge workers who want to be reliable, not just fast. It gives you one principle, one loop (Discover, Screen, Extract, Synthesise, Verify) and one real project to practise on.
Ten chapters
- Why AI Matters for Research
- The AI Research Tool Landscape
- Research Techniques
- Building Your AI Research System
- Prompting for Research Work
- Organising Knowledge
- Collaboration with AI
- Ethics, Disclosure and Accountability
- Choosing and Evaluating Tools
- Staying Current
How it was edited: the first edition was critically reviewed, then rewritten to remove every unsupported statistic. For the second edition, four AI models reviewed the manuscript independently. I merged their reviews and kept the byline, as the Claim Rule says.
Also on the bench
The rest of the lab.
-
Building
NeuralHub
A shared knowledge vault for teams. The AI assistants a team already uses can read from it and write back to it over MCP, so everyone works from one source of truth.
-
Live
SID Desk
An after-hours lead agent for tradespeople. It answers the enquiry, qualifies the job and pings the owner, so no lead waits until morning.
See it -
Research
Trading Arena
Three AI agents race each other in paper trading. My main finding so far is that the prompts, not the models, cause 93–99% of “hold” decisions.
-
Concept
Salvage
A tool that captures each AI build as a runbook (the code, the how-to and the gotchas that cost hours) so the next session on any platform can pull it back up.
-
Daily
The daily briefing
An overnight agent reads AI, GitHub, markets and new-tech sources and publishes a short briefing at 02:00 Dublin time.
Read today's
Lab notes, when there's something real.
I send an email when something ships, along with what broke on the way and how I fixed it. No schedule, no filler.