What we’ve built.
Client projects and our own products, labelled so you always know which is which. Where something is still in progress, it says so.
Built for other people
Work someone paid for. Live links where the site is live.
aROMA Laundry
Shopify storefront for an Italian laundry scent-booster brand selling into Canada.
What we did: Built the store, then audited it with our own SEO engine and worked through the findings.
Ageing Gracefully
Site for an independent Pilates, yoga and functional-strength studio in Brampton, Ontario.
What we did: Design and build, structured around the classes, memberships and local searches people actually make.
PB Renovations & Restorations
Rebuild for a custom millwork, cabinetry and restoration shop in Sarasota, Florida.
What we did: Full rebuild replacing a page-builder site, with the services and portfolio restructured around what buyers search for.
Systems, not sites
Our own research and tooling. None of these were paid engagements, and we describe what they do rather than how they are built — the implementation is the asset.
Private AI infrastructure
AI context proxy
A layer that sits in front of a language model and cuts down what it has to read before every request, without the application calling it knowing anything changed.
What we did: Built it and measured it — roughly a third fewer tokens on long conversations with output quality held steady.
What that means we can build: Bring down what your AI usage costs as it scales, without rewriting the product around it.
Private model hosting
Run, manage and serve your own language models on your own hardware — download, swap and retire them from a simple interface, with no third-party service in the path.
What we did: Built the serving layer and the management interface, and tested a range of models and inference engines against real workloads.
What that means we can build: Use AI where the data legally or commercially cannot leave your building.
Search over your own documents
Ask questions of your own material and get answers with the sources attached, instead of a confident guess from a public model that has never seen your files.
What we did: Built the ingestion and retrieval pipeline end to end, running entirely on local infrastructure.
What that means we can build: Give staff straight answers out of your contracts, manuals and history — with citations they can check.
Private AI workspace
A company's own AI chat platform: staff accounts, conversation history, an admin view of who is using what, and a mode where the assistant can actually run tasks rather than only talk about them.
What we did: Built the whole platform — sign-up, chat, administration, usage monitoring and the tool-running mode — to deploy on infrastructure the company controls.
What that means we can build: Give a company its own private assistant, with accounts and usage control, instead of paying per seat for someone else's.
Training data pipeline
Turning raw scraped web content into material a model can actually learn from — cleaned, de-duplicated, analysed and split into coherent pieces.
What we did: Built the pipeline end to end, for both retrieval and fine-tuning use.
What that means we can build: Get your own messy source material into a state where an AI system can use it reliably.
Computer vision
Drone detection and tracking
Real-time detection and multi-object tracking from aerial video — following each object across frames, holding identity when it is briefly lost, and reading movement and direction.
What we did: Trained the detection models on custom imagery and built the tracking and trajectory analysis around them.
What that means we can build: Turn a camera feed into structured events something else can act on.
Chart pattern detection
A desktop application that watches live financial charts on screen and flags recognised patterns as they form, with adjustable sensitivity and a timestamped log of everything it saw.
What we did: Trained the detection model on chart imagery and built the application around it, packaged so it runs on a machine with nothing else installed.
What that means we can build: Watch a screen or a feed for something specific and record every time it happens, without a person sitting there.
Data and decision support
Customer segmentation and churn risk
Groups a company's customers by how recently they bought, how often, and how much they spend — separating the loyal from the slipping from the already gone — and puts it in a dashboard a sales team can actually read.
What we did: Built the analysis, trained the model that places a new customer into a segment, and built the dashboard around it.
What that means we can build: Tell you which customers are about to stop buying while there is still time to do something about it.
Autonomy and field systems
Natural-language ground control
A control station where you direct drones and swarms by typing what you want, and the map itself gives the AI the context to interpret it — devices, positions, zones and geometry.
What we did: Designed and built the whole thing, including the reasoning layer that turns an instruction into a safe, specific action.
What that means we can build: Put a plain-language layer over hardware and live operational data, running entirely on machines you own.
Off-grid mesh messaging
Passing messages across a long-range radio mesh with no internet, no cell service and no infrastructure in between.
What we did: Built the relay that streams generated messages across the mesh and out to handsets in the field.
What that means we can build: Keep people and systems talking where there is no connectivity to rely on.
Fleet telemetry simulator
Simulates a fleet of drones reporting position and sensor data across a network, with a live dashboard showing every message as it arrives.
What we did: Built the simulator and the dashboard so software can be developed and tested before any hardware exists.
What that means we can build: Stand up a realistic test environment so the software is ready before the hardware arrives.
Learned motion control
A trained model that works out the joint positions a robot arm needs to reach a point in space, paired with a simulation to check the result before it moves.
What we did: Generated the dataset, trained the model and built the simulation to validate it.
What that means we can build: Train a model to control physical motion where the exact maths is expensive, unknown or too slow.
flogtool
An open-source analyzer for drone flight logs.
What we did: Built and published it.
What that means we can build: Make sense of machine telemetry after the fact.
Search and web tooling
SEOGen
An audit engine that measures how visible a site is to search engines and to AI answer engines — every finding a real check against the live site, not a model's opinion.
What we did: Built it, and use it on client work. The test suite runs 159 cases with no network access, so the results are repeatable.
What that means we can build: Tell you exactly why you are invisible online, with evidence rather than adjectives.
Flaitera
A productized web-improvement service: measure, ship, document, repeat — with the improvement log published in public.
What we did: Built and running it as our own experiment in turning the work into a product.
What that means we can build: Turn a repeatable service into something that runs on a subscription.
Five systems, five registers
A design system defines colour, type, spacing, components and voice once, so every page after the first takes hours instead of days. These were built for five very different industries — the range is the point. Open any of them: they are full working documents, not mockups. We build these for clients, and they are yours to keep.
Datum Aerospace
Built like a controlled document: revisions, approvals and test conditions on the page, because customers who qualify hardware for flight are persuaded by traceability, not adjectives.
Open the full system →
Datum Civil
Treats every screen like a drawing sheet — fixed grid, title block, dimensions you can trust — so the site feels as rigorous as the stamped set.
Open the full system →
Marlow & Reeve
Quiet, deep and specification-grade, built to feel the way a hand-rubbed lacquer feels. Restraint as the proof of quality.
Open the full system →
Scribble
A UI kit drawn from one pencil line — wobbly boxes, hand-lettered type and stick figures, so a product can feel approachable without buying an illustration pack.
Open the full system →
FloLogix AI
Our own. Cyan carries, yellow signals, and a single seven-degree tilt taken from the logo runs through everything. This site is built on it — which is why a new page takes hours.
Open the full system →Want something like this?
Thirty minutes on a call and you will have a written scope and a fixed price.