Clients and project names are held back. Everything else is real: what the problem was, what got built, what changed as a result. Where a number has not been measured or cleared, it is marked instead of guessed.
01 / Vision
Live movement tracking
Fitness technology Multi-camera floor, two views per station
Running in production
Problem
A training floor has more happening on it than any coach can watch.
Sector
Fitness technology.
Built
Every station watched continuously, from paired camera views, with no wearable on the member.
Reps counted and depth measured from the movement itself, rather than from a timer or a button press.
A stall detector that recognises when someone has stopped being able to finish a rep, and says so while it is still happening.
One independent process per station and movement, so a failure at one station never takes the floor down.
An event layer that survives a dropped connection and reconnects without a restart — the difference between a demo and a system that opens at six every morning.
Rebuilt for production speed, and extended with muscle-activation sensing for the sessions that need it.
Outcome
Coaching attention goes where it is needed. The floor measures itself — every rep, its depth, and the moment a member stalls — so a coach walks the room instead of counting it, and a member’s progress is a record rather than a recollection.
02 / Vision
Equipment recognition by weight
Fitness technology Twenty kinds of equipment
Live alongside 01
Problem
Knowing someone picked up a dumbbell is not enough. The system has to know which one.
Sector
Fitness technology. Same floor as 01.
Built
Twenty kinds of equipment told apart in production — dumbbells from 5 to 50 lb, six kettlebell weights, four colour-coded balls — grown from an earlier three.
Equipment separated by weight, not just by type: a 20 lb dumbbell and a 25 lb dumbbell are different answers, not the same one.
A colour classifier for the weighted balls, reading load off colour alone, since the balls are identical in shape and size.
Learned from examples rather than set as a fixed colour rule, because gym lighting shifts colour far enough to break one.
Answers returned to the station in the same moment the movement is being tracked, so weight and rep arrive together.
Outcome
A session logs itself. The load a member actually lifted is captured without anyone typing it in, which means progression is measured from what happened on the floor rather than from what someone remembered afterwards.
03 / Vision
Vehicle damage from a phone
Automotive On-device detection and depth
Built and iterating
Problem
Assessing a vehicle’s damage takes an expert and a clipboard. It should take a walk around the car.
Sector
Automotive.
Built
Thirteen kinds of damage, learned from thousands of photographs of real vehicles rather than from staged examples.
Detection running live on the handset itself, with no round trip to a server and no signal required.
Depth sensing that reads the shape of the panel, not just the picture of it.
A dent detector using two independent depth methods, so a shadow does not read as a dent.
Close inspection mode, a report generated at the end of the walkaround, and a capture mode that turns each scan into training data for the next model.
Outcome
An assessment that needed an expert now needs a phone and two minutes. Every vehicle is scored the same way by the same model, so the report does not change with who is holding the device — and the system gets better each time it is used.
04 / Vision
Contactless vitals sensing
Consumer health Camera and radar sensor fusion
Spec and hardware prototype
Problem
Parents want to know a sleeping baby is breathing without putting anything on the baby.
Sector
Consumer health.
Built
Three sensing methods fused into one reading — camera, radar, and room conditions — so no single sensor has to be right on its own.
A working monitor reporting heart rate and breathing rate across a room, with nothing attached to the child.
A companion phone app for the live view.
A regulatory line drawn through the feature set at design time, separating what can be claimed from what cannot.
Outcome
Vitals read across a room with nothing touching the child, and a product that can ship without waiting on a clearance path — because the line between a wellness claim and a medical one was drawn before the hardware was built, not after.
05 / Vision
Body composition from vision
Fitness technology Camera capture and pose landmarks
Running in production
Problem
Measuring body composition normally means a scale, a caliper, or a scanner the size of a wardrobe.
Sector
Fitness technology.
Built
Composition estimated from camera capture and body landmarks — nothing worn, nothing touched, no dedicated booth.
What the app and the measurement side promise each other, settled and versioned before either shipped, so each can change without breaking the other.
Capacity measured under real conditions rather than assumed, with the numbers written down.
Validated against a reference method before release, not after.
In production and serving real users.
Outcome
A measurement that used to need equipment and an appointment now needs a camera and a few seconds, which means it can be repeated often enough to show a trend rather than a single reading.
06 / Vision
VisionGuard
Face recognition at the door Healthcare and secure sites On-device, watchlist-only
Built and running on device
Problem
A facility knows who is credentialed. It does not know who just walked in.
Sector
Our own product. Healthcare and secure sites. Entrances, reception, and loading doors, where a badge says someone should be allowed in but nothing says the person at the door is a barred visitor or a former employee.
Built
One-to-many matching against a single facility’s own watchlist — approved vendors and flagged individuals — and nobody else. Everyone else is simply not a match: no profile, no record, nothing kept.
A face becomes a faceprint — a numeric signature, like a fingerprint — computed on the device itself. The photograph never leaves the building.
Detection and matching run on the device itself, on both kinds of phone, so recognition never waits on a server.
A match is a prompt for a person to confirm or dismiss. The system never acts on a face by itself.
Built on our own recognition system rather than a licensed one, so it carries none of the restrictions that come bundled with something bought off the shelf.
Outcome
The security gap closes without opening a privacy one. A facility recognises the people who matter at its doors — the approved vendor waved through, the barred visitor flagged while they are still standing there — without profiling everyone else who walks past. Because matching happens on the device against a defined list, the sensitive data mostly never exists to be leaked, which is a stronger guarantee than any data-handling clause. In a hospital that is the difference between a system that can be deployed and one that cannot.
07 / Signal
Call centre conversations
Contact centre operations Live calls and archive, real time
Delivered
Problem
A floor manager hears a handful of calls a week and has to run a whole team on that.
Sector
Contact centre operations.
Built
The whole recorded archive turned into text, alongside live calls, so what was only ever stored became something you can actually search.
Emotion recognised from the audio itself — how something was said, not only what was said — running live as the call happens rather than inferred from a transcript afterwards.
The turn in a conversation marked at the moment it happens, so a call going wrong is visible while it can still be saved.
Script adherence checked on every call rather than on a sample of them.
Findings surfaced to the agent and the manager while there is still time to act on them, not in a report a month later.
Outcome
Coaching stops being a sample, and it stops being after the fact. Every call is reviewed instead of the few a supervisor had time for, so the agents who need support are found by what happened on the line rather than by who happened to be listened to that week — and because the emotion in a voice is read while the call is live, a conversation going wrong can be caught during it, not in a report the following month.
08 / Signal
Policy classification
Healthcare Unstructured document corpus
Delivered
Problem
Years of policy documents, more than any team can read, hiding product opportunities nobody could see.
Sector
Healthcare.
Built
Documents classified by what is inside them, not by where somebody filed them.
Text mining across the entire corpus, surfacing the patterns nobody had the hours to find by hand.
Delivered as a cloud service the client’s own team runs on new documents, rather than a one-off analysis that goes stale.
Outcome
Unstructured documents became a product input. The client could see which customer needs kept recurring across the whole corpus and build against them, instead of guessing from the handful of documents someone had time to read.
09 / Signal
Dereal
A real-time voice companion Voice AI and conversational systems Full duplex, private by design
Built
Problem
A person says something heavy, out loud, and expects to be met — not lectured, not stalled, not handed a five-step plan.
Sector
Mental health and wellbeing. Our own product. Anyone can wire a language model to a microphone; that is the demo. The line between a companion and a liability is drawn in the parts nobody screenshots — what happens in a crisis, what happens when the model wanders out of its lane, and what happens to the conversation after it ends.
Built
Full duplex, so it is a conversation rather than a transaction: you speak, it listens, thinks and answers in real time, and interrupting it mid-sentence stops it, the way interrupting a person does.
Seven companions, each a different way of being met — life direction, habit and behaviour, insight, emotional and somatic healing, relationships, memory work, and creative purpose. Not seven prompts wrapped around one voice; seven genuinely different conversational presences, and the system moves between them as the conversation moves rather than making someone choose, announcing the handoff so nobody is left wondering why the voice changed.
Grounded in eight evidence-based clinical frameworks rather than an invented bedside manner — cognitive behavioural, dialectical behaviour and rational emotive behaviour therapy, schema therapy, somatic experiencing, mindfulness-based stress reduction, trauma-informed care and motivational interviewing. What a companion reaches for, and when it stops pushing, comes from those rather than from a prompt written to sound caring.
Every turn read for distress before a reply is formed. When it finds it, the companion slows down, stays present, stops probing, and surfaces the resources that matter.
More layers run on a turn before a word comes back. Alongside that safety read, one settles the register — hold space, answer plainly, name a pattern, push back on self-attack, or witness without probing — one picks the therapeutic grammar the answer is built in, and one tracks whether the session is opening, working or closing.
Then the answer is checked against what was asked for, and sent back if it missed: a trailing question stripped when someone wanted a direct answer, a reply regenerated when it drifted into clinical jargon, turned metaphorical during a crisis, or refused authority the companion is meant to hold.
A boundary that holds. It is a wellbeing companion, not a therapist, a lawyer or a business coach. Asked to grow someone’s sales it says that is not its lane and turns back to the person — and it keeps doing that as the conversation drifts.
Each person’s data isolated and access-checked, encrypted at rest, with an audit trail of who reached what.
Memory that carries between sessions — commitments someone made, topics that keep returning, patterns that only show up across weeks. The next conversation opens by asking how the thing they said they would try went, rather than from nothing.
Speech in, understanding, and natural voice out — the whole path runs inside our own environment, so no third-party service ever holds the conversation.
A second product for the practitioners: Dereal Pro. A facilitator dashboard over client progress between sessions, alerts when the system detects distress, and a hard boundary between personal and clinical data — what someone says to a companion does not quietly become a case note. It is built to augment a therapist, not to stand in for one.
Outcome
It held real conversations. The part that took the work is not the voice loop but everything guarding it: distress recognised before a reply is formed rather than after, a lane the model stays inside as the conversation wanders, and a pipeline that never hands a person’s worst moment to someone else’s service. Those are the parts that decide whether a wellbeing product can be put in front of people at all.
10 / Signal
Intellinova Law
Legal services Our own product Self-hosted, single tenant
Built and deployed
Problem
A practice runs on documents nobody can search, drafting nobody has time to review, and billing that has to satisfy a tax authority.
Sector
Legal services. Our own product, built end to end rather than assembled from a vendor’s parts.
Built
Everything goes through one common entry point. The web app holds no privileged path of its own — a second application, ours or a partner’s, works from exactly what the browser does.
The practice’s own documents indexed for retrieval, so they can be asked a question rather than searched by filename.
Drafting, work-product review and risk flagging — where the work is in finding the right passages and in the review around what comes back, rather than in the model itself.
Invoicing that carries the QR a tax authority requires, generated on request and never stored.
Every file behind one storage boundary from the very beginning, so moving the data to another provider, or into a particular jurisdiction, is a settings change rather than a rebuild.
Deployed single tenant, with the application, its records and its files under one roof, and the whole arrangement written down so it can be stood up again exactly rather than from somebody’s memory.
Outcome
Built and deployed, and waiting on its first real matter. The decision that will matter most is the storage boundary: because every file passes through one storage boundary, the question a practice always asks — where does this data physically sit — is answered by configuration instead of by a rewrite. For a firm choosing whether it can adopt a system at all, that is usually the question the answer turns on.
11 / Signal
DeepMinds
Document intelligence Our own product Teams, roles and seats
Built
Problem
An answer pulled out of a hundred documents is worth very little if nobody can see where it came from.
Sector
Document intelligence. Our own product, built as a platform rather than for one industry, then tuned for several.
Built
Answers carry their sources. Every claim in a reply is numbered inline, hovering shows the document, the page and how close the match was, and clicking jumps to the exact text highlighted in the original — so the answer is checkable rather than trusted.
Tables lifted out of documents and kept as figures in their own right, so a question about numbers is answered from those figures rather than from prose that happens to mention them.
A reader’s own notes folded back into retrieval. Annotations are indexed alongside the source material, so the system answers with what this person already worked out, not only with what the documents say.
Questions asked across a whole shelf at once rather than one file at a time, with the documents organised in a hierarchy instead of a flat list.
Built for teams from the outset — organisations, four levels of access, seat limits and invitations — rather than a single-user tool that had sharing added to it later.
The same engine tuned per profession, so the prompts, templates and vocabulary differ for a legal, medical, financial or academic reader working on the same kind of file.
Reading a new upload happens on its own track, so a large one being taken in never blocks somebody waiting on an answer.
Outcome
What decides whether a tool like this gets used twice is not the answer, it is the citation. A reply that cannot be traced back to a page is an opinion, and professionals do not act on opinions from software. Bringing the retrieval, the extracted tables and the reader’s own notes into one answer — each part of it traceable to where it came from — is the difference between a system somebody works with and one they try once.
12 / Data
Growth drivers
Subscription business Internal and external data
Delivered
Problem
Subscriber growth had stalled, and the data on hand could show that it happened but never why.
Sector
Subscription business.
Built
External data brought in alongside the client’s own, because the answer was not sitting in the internal numbers.
Modelling that separates the indicators which lead growth from the ones that only follow it.
Every indicator scored by direction — helping, hurting, or doing nothing at all.
Outcome
The growth conversation moved from opinion to evidence. The team could tell which levers genuinely moved subscriptions from the ones that only looked like they did, and spend effort accordingly.
13 / Data
Cross-site scheduling
Healthcare Multi-site surgical network
Architecture and prototype
Problem
Each site schedules alone, so one theatre runs late while another sits empty.
Sector
Healthcare. Networks operating several surgical facilities under shared governance.
Built
Case length predicted from what actually happened before, replacing block estimates that were set once and never revisited.
Cancellation risk scored ahead of time, so a slot can be refilled before it is lost rather than after.
Demand forecast by specialty and by facility, feeding capacity and staffing decisions weeks out.
Placement across sites solved as one problem instead of several — the piece the incumbents do not have.
Full data model, an integration layer for existing clinical records, and a working prototype the network can steer with.
Outcome
A network can see its whole capacity at once. Where a case should go stops being whichever site the referrer knows and becomes a decision made against real utilisation, real travel time, and real equipment fit.
14 / Data
A cloud data lake, and the workflows around it
Healthcare compliance Data lake, process automation, analytics A programme, not a project
Delivered
Problem
Years of operating data spread across systems that disagreed with each other, and no route from any of it to a decision.
Sector
Healthcare compliance, at enterprise scale.
Built
The shape all of it needed to end up in, agreed before anything was built, so what got built answered where the business was going rather than fitting itself around what was already there.
One place in the cloud to hold all of it, designed and then tested against real volumes with the results written up — and sized on a trial run first, so the capacity the business committed to had already been measured rather than guessed at.
A plan for what data gets collected, from which system, and in what order — the question of sequence that decides whether the thing fills up or sits empty.
One agreed definition of each thing the business tracks — a site, a contract, a person — and a list of them everyone works from, so the same word means the same thing everywhere. Without that, all anyone has built is somewhere to keep the disagreements.
The reporting and analysis work carried from first plan through four revisions to delivery, rather than a strategy document handed over at the end.
Every business process drawn end to end before any of it was automated, the sales process included. Automating a process nobody has mapped only makes a bad one run faster.
The system that would run those processes chosen through a written case and a scored comparison, then stood up small and actually run before anyone committed — judged on how it behaved, not on how it read on paper.
How work flows and how the systems fit together designed as one, so joining a process across several systems did not quietly rebuild the single large system they were meant to replace. Standards written alongside, so teams worked to one definition rather than each inventing their own.
Every choice made in the open, each option scored on a comparison the client kept, so the reasoning outlived the engagement.
The business mapped before the systems were: what it is able to do, how it runs day to day, and the customer’s path from end to end — so the data work answered questions the business actually had rather than the ones the data happened to make easy.
Delivered over several years and several parallel efforts, with the design, the standards and the reasoning behind each choice handed over as documents the client’s own teams carried on from.
Outcome
The business moved from data it stored to data it could ask questions of. The work that decides whether that happens is unglamorous and mostly invisible — agreeing what a thing is called, deciding what gets brought in and in what order, proving it holds up under real load — and it is the difference between an analytics programme that delivers and one that produces a roadmap nobody executes. The design, the standards and the reasoning behind each choice stayed with the client rather than leaving with us.
15 / Process
Process automation at enterprise scale
Government, healthcare and large enterprise Business process management Across several organisations
Delivered
Problem
A large organisation runs on processes nobody has ever drawn. Work moves by email, habit and whoever remembers the exception, and when it stalls nobody can say where.
Sector
Government, healthcare and large enterprise operations. The work is the same shape in each: many hand-offs, real consequences for getting one wrong, and an audit trail somebody will eventually ask for.
Built
The process drawn end to end first, with the people who actually run it, before a line of it is automated. Automating a process nobody has mapped only makes a bad one run faster, and at enterprise scale it makes it faster in every department at once.
The choice of system made in the open — a written case, scored against the organisation’s own criteria, then built small and run. Chosen on how it behaved, not on how it read on paper.
How work is joined up designed alongside how the systems are built, so a process crossing six of them does not quietly become one big new system with a workflow engine sitting in the middle.
The exceptions treated as the design problem they are. The happy path is the easy part; what a process does with the case that does not fit decides whether people use the system or route around it.
Approvals and hand-offs set out explicitly, with who decides and what they see at each step — which is what makes the trail auditable afterwards rather than reconstructed.
Ownership and standards written down, so a process has someone accountable for it and teams work to one definition instead of each drawing their own.
Measured at every point where work waits, because the queue nobody measures is the one the organisation has quietly built its staffing around.
Outcome
An organisation that can see its own work. The value is rarely the automation itself — it is that a process which existed only in habit becomes something visible, owned, measurable and changeable. Once a queue can be seen, the argument about where the bottleneck is stops being a matter of whose department is loudest, and the exceptions stop being invisible work absorbed by whoever has been there longest.
16 / Process
The journey before the system
Enterprise services Customer journey and operating model Mapped end to end
Delivered
Problem
Every department could describe its own part of the customer’s experience. Nobody could describe the whole of it, which is where the experience was actually going wrong.
Sector
Enterprise services. An organisation large enough that no single person had seen a customer’s path from first contact through to renewal.
Built
The customer journey drawn end to end, across every department that touches it, as one picture rather than as each team’s view of their own segment.
Every point of contact identified and described the same way, so a hand-off between two departments could be discussed as a single event instead of as two different events with two different names.
The commercial process mapped alongside it, because a journey drawn without the sales path describes what the organisation intends rather than what a customer meets.
What each part of the business is able to do, and how it runs, set against that journey, so it was clear which part owned each stage and which stages had no owner at all — usually the ones customers complain about.
The principles, the goals and the pressures behind the work written down first, so later arguments about priority had something to be settled against other than seniority.
The result kept as living documents the client’s own teams maintained, not a slide deck presented once and filed.
Outcome
The organisation could see its own customer’s path in one piece. That sounds modest and is not: most experience problems in a large business are not inside any department but in the seams between them, and a seam is invisible to everyone whose map stops at their own boundary. Drawing the whole journey is what turns “our customers are frustrated” into a specific hand-off, owned by someone, that can be fixed. It is also the work that decides whether the automation which follows is aimed at the right thing.
Bring us the hard half of the problem.
A feed nobody has time to watch, or a pile of documents nobody has time to read.
Smart Waves