Some of this is our own product work, which we can describe fully. Some is client work described at sector level, because those clients asked not to be named — and we would rather show you less than name someone without permission.
A private messaging platform built so that the operator cannot read the messages either.
A therapeutic game designed around cognitive health and relaxation rather than engagement metrics.
An infinite arcade game built as a stress test for browser-based real-time rendering.
Vessel management and compliance for a ship management group, built to work with no connectivity.
A confidential case intake and management platform for a firm handling fraud and cyber-crime matters.
Smart contracts, distribution and public interface for a digital asset launch.
Storefronts, marketplaces and the fulfilment integrations behind them, across several retail clients.
Where a client asked not to be named, the project is described at sector level. We publish no metric we cannot substantiate, and no client logo we do not have written permission to use.
Original product directions from our own R&D — not client engagements, not delivered work. We publish them because how a team thinks about a problem is more useful than a logo wall.
An answer engine over an institution’s own records that never leaves its building.
Open-weight models running entirely on-premise, indexing internal records with document-level access control carried through retrieval, so an answer can never cite a document the asker is not cleared to read. Every response carries its sources and a confidence signal, and low-confidence questions route to a human rather than being guessed at.
Provenance for temperature-sensitive cargo that a receiving party can verify independently.
Low-power sensors log temperature and location continuously, signing readings on-device so they cannot be edited after the fact. Summaries are anchored on-chain at custody handover, giving the receiver a record they can verify without trusting the shipper — the narrow case where a distributed ledger genuinely earns its cost.
High-risk procedures rehearsed in VR with assessment that means something.
A virtual reproduction of a specific plant and its equipment, where operators rehearse hazardous procedures with realistic failure modes. Assessment measures sequence, timing and decisions under induced fault conditions, producing a competence record an auditor can inspect rather than a completion tick.
Distribution losses detected as they begin rather than at the next audit.
Meter telemetry modelled per feeder to learn normal consumption shape, flagging divergence that indicates loss, tampering or developing equipment fault. Detections are ranked by estimated financial impact so field crews are dispatched to what matters, and every flag carries the evidence that produced it.
Beneficiary records that survive a raid, a seizure or an outage.
Field-collected records encrypted on-device and sharded across jurisdictions so no single seizure yields readable data, with duress codes that present a decoy dataset. Built for organisations working where the safety of the people in the database is the threat model, not an abstraction.
Know which of your controls a regulation change actually affects.
Continuous monitoring of regulatory sources, mapped against an organisation’s documented controls, so a published amendment produces a specific list of affected policies and systems rather than a PDF someone must read. Every mapping is traceable to the clause that produced it, because an unexplainable compliance recommendation is worse than none.
Atomos Labs concepts are internal research directions. They are not client engagements and have not been delivered. If one of them describes a problem you have, we would be glad to talk about building it properly.
Most of our work runs under NDA. Tell us what you need built and we will tell you honestly whether we are the right team for it.
