Our Work
Showcase
Developed for real users — fully instrumented, production-grade software, not prototypes. From Flutter apps live on iOS and Android to an AI-native personal operating system: complex algorithms, engagement mechanics, Cloud Functions pipelines, subscription management, and zero crashes.
D4L — Design For Life
A personal operating system — R&D
Our research platform, and the clearest statement of how we engineer. D4L captures one household's entire record — email, documents, calendars, health data, app data — into a sovereign, privately owned knowledge graph, then governs the decisions on top of it: an attention engine, a policy layer that acts only within rules you've ratified, and an AI interface bound to propose but never legislate. The hard part isn't the AI; it's the trust architecture that makes it safe to run over a life — and because the record speaks an open protocol, every time frontier AI gets smarter, the system does too.
- Ask your life anything — "when did the symptoms start?", "what did I promise people and what's overdue?", "which insurance was active that day?" — answered across decades with the sources attached
- Two append-only ledgers are the only source of truth; every index, graph, queue and live situation is disposable and rebuilds from them exactly — the past stays queryable as of any date
- A decision layer reads what each event means, matches it to policies you've ratified, and takes only reversible actions on its own — anything that sends or deletes waits for you; it first resolves whose job an item even is — yours, the other party's, shared or unknown — and new behaviour runs in shadow for weeks before it can change what you see
- The AI proposes, only a human legislates: it drafts new rules in plain language and changes nothing until you ratify — decisions are never minted by a model, advice is evidence-gated
- 70+ architecture decision records, 140+ forced-failure acceptance suites, and an assurance layer that independently proves each control actually ran in production — not just that its tests passed; deterministic replay, zero personal data in the repository
My 12-Step Coach
Recovery structure in your pocket
If your idea depends on users forming a habit, this is the reference. My12SC is a behaviour change platform — the hard part wasn't the content, it was engineering the engagement mechanics that keep people coming back.
- Onboarding wizard with complex state management — turns setup into an experience that predicts long-term retention
- Full habit-loop system: streak mechanics, FCM push nudge timing, dormant-user re-engagement, gamified milestone rewards
- Insights engine with progressive reveal — complex data surfaced without overwhelm
- Subscription management, home screen widgets (iOS WidgetKit + Android Glance), sharing mechanics
- DAU/MAU above peer median — Zero crash rate — Firebase Analytics with full cohort segmentation
Surplus
Financial clarity for households
If your idea involves data, calculations, or intelligence — this is the reference. Surplus is a fintech platform with a custom algorithmic insights pipeline, AI integration, and a learning categorisation engine — its v1.1.0 release rebuilt that intelligence layer from the ground up.
- Custom Buffer engine: forward-looking financial runway in days, calculated across all accounts and committed obligations
- Rebuilt insights engine: recurring payments identified from the bank reference itself — direct debits and standing orders no longer hide behind cryptic strings — surfacing locked income and upward cost drift
- Every signal declares its own confidence — Active when the data is ready, Gathering while it builds — so the UI never shows false precision
- Calibration layer plus a learning reclassification engine: user input always outranks inference, and corrections train a persistent merchant mapping that sharpens every future import
- AI API integration for OCR statement capture, real-time household sync, home screen widgets — Zero crash rate in production
The Smarts Family
Better taste, less money — from your own data
If your idea needs real intelligence — not a generic model bolted onto a database — this is the reference. Three apps, one philosophy: a user's own ratings reveal what actually drives their taste, and everything is ranked by value for money. The point is what sits underneath — we don't reuse one model, we re-architect the intelligence layer around how each domain actually works.
- Coffee Smarts — the taste drivers are printed on the bag (variety, process, origin, altitude); a structural Ridge regression finds which of them move each user's ratings
- Whisky Smarts — the drivers hide below the label, so a custom causal flavour-vector model separates intensity from character: two peated drams a drinker loves and hates aren't a contradiction
- Olive! — label-scanning captures the data most people can't read, then a cultivar/origin/intensity engine surfaces what's behind their best bottles
- Shared across all three: €/★ value rankings, a plain-language palate read, and a reference encyclopedia overlaid with the user's own collection
- Each built on the Coffee Smarts foundation — concept to live on both stores in weeks, fully instrumented, zero crashes
NoiseMark
Measure it. Mark it. Keep the record.
If your product has to produce evidence somebody else will accept, this is the reference. The measurement was never the hard part — the documentation was. NoiseMark is built so the record writes itself, and so that every claim it makes can be defended.
- Real-time audio DSP on the device — A-weighted, Fast, one-second energy buckets, validated against a reference meter across the full working range
- An explicit separation of measuring from saving: a rolling in-memory buffer, and nothing written to disk until the user deliberately acts
- Crash-durable capture — at most two seconds of history is unrecoverable after process death, proven by test, not asserted
- Reports that state their own limits: measured time versus wall-clock, interruptions shown, no double-counting across overlapping records
- Privacy by construction — no audio ever stored, records held on the device with optional Cloud sync and backup to the account, and a documented open archive so years of records survive a new phone
