Applied AI
Perception, multimodal systems and decision support, built with the evaluation harnesses that prove they behave.
- Computer vision & sensor fusion
- LLM & multimodal systems
- Evaluation, red-teaming, drift monitoring
- On-device & edge inference
Product innovation consultancy
We turn frontier technology into products that ship.
Sirotics works at the edge of applied AI, spatial computing and robotics, then carries that work all the way through research, design, engineering and clinical validation, into markets where being interesting isn't enough.
Scroll
Applied AI · Spatial computing · Robotics, brought through product research, design and engineering into digital health, medical devices and clinical research.
ISO 13485
IEC 62304
ISO 14971
IEC 62366 human factors
21 CFR Part 11
HL7 FHIR
ROS 2
OpenXR
Edge inference
IEC 60601
CDISC SDTM
ISO 27001
WCAG 2.2
Zephyr RTOS
ISO 13485
IEC 62304
ISO 14971
IEC 62366 human factors
21 CFR Part 11
HL7 FHIR
ROS 2
OpenXR
Edge inference
IEC 60601
CDISC SDTM
ISO 27001
WCAG 2.2
Zephyr RTOS
How we're built
Three layers,
one team.
Most firms hand you one of these. Deep technical capability with no route to a product, or a tidy product process with no real engineering underneath it.
We run all three together, because that is the only way emerging technology actually reaches a user, and survives a regulator.
Layer 01
Practices
The technical depth
Small, senior teams who have shipped in these domains before. They run feasibility, de-risk the hard parts early, and stay on the programme through production rather than handing over a report.
Layer 02
Product studio
The delivery engine
Research, design and engineering working as one group on a single backlog. Industrial design sits next to firmware; human factors sits next to the requirements that verification will be written against.
Layer 03
Regulated health
Where the bar is highest
Our deepest sector. Quality, risk and evidence are designed in from the first sprint, not retrofitted six months before submission, which is where most programmes lose a year.
Capabilities
Eight practices that
work on one programme.
You can engage any one of these on its own. Most clients come for one and stay for three, because the handoffs between them are where programmes usually break.
See capabilities in fullSpatial computing & XR
Immersive systems for training, simulation and therapy, built for real headsets and real users, not demo loops.
- VR / AR / MR application development
- Simulation & digital twins
- Procedural & surgical training
- Haptics, comfort & interaction design
Physical AI & robotics
The next front in the AI race. Perception, learning and control for machines that act in the real world, sensors through to deployed fleets.
- Sensing, sensor fusion & dynamic vision sensors
- Embodied AI, robot foundation models & sim-to-real
- Drones, mobile robots & manipulators
- New robot R&D, automation & fleet deployment
Product research
Before anything gets built: who it's for, what it has to do, and whether it can actually win.
- Generative & ethnographic research
- Human factors (IEC 62366, FDA HFE)
- Technology & IP landscaping
- Business case & roadmap strategy
Product design
Industrial design, interaction design and design systems that survive contact with manufacturing.
- Industrial design & CMF
- UX / UI & design systems
- Looks-like / works-like prototyping
- Accessibility & inclusive design
Product engineering
Electronics, firmware, mechanical and software under one roof, through DFM and into production.
- Electronics & PCB design
- Embedded firmware & RTOS
- Mechanical, DFM & tooling
- Cloud, mobile & systems engineering
Digital health
Connected products in regulated environments, software as a medical device, wearables, diagnostics and the platforms behind them.
- SaMD & digital therapeutics
- Wearables & remote monitoring
- Interoperability (FHIR, DICOM)
- Cybersecurity, HIPAA & GDPR
Clinical trials & evidence
Technology that makes studies actually run, and evidence that holds up with regulators and payers.
- Decentralised & hybrid trial platforms
- eCOA / ePRO, eConsent, wearable endpoints
- Digital biomarker development
- 21 CFR Part 11 & GxP validation
AI research
A research practice,
not a research slide.
Anyone can build an AI demo in a fortnight. Almost nobody can turn one into a system that runs unattended, behaves predictably on their own data, and holds up in front of a customer.
We run active research on the things that actually block that: memory and long-context systems, agentic architectures, LLM adaptation, in-house data extraction, evaluation science, on-device efficiency, and enterprise adoption, where most programmes quietly stall.
It produces things that leave the building.
Kerneta, the open .dai memory format behind DaiDocs, began as an internal answer to a client problem and is now a Sirotics spinout. It ranks 2nd on LongMemEval-S, the AI memory benchmark, at a third of the tokens of the system above it.
Sectors
Where this work
lands.
We go deep in a small number of markets rather than shallow across all of them. The constraints in these sectors are specific, and knowing them in advance is most of the value.
All sectors
01
MedTech & diagnostics
Class I–III devices, point-of-care instruments and the software that runs them, from concept through V&V and transfer.
02Digital health & therapeutics
SaMD, prescription digital therapeutics and remote monitoring, built around adherence and evidence, not just features.
03Life sciences & clinical research
Decentralised trial technology, digital endpoints and lab automation for sponsors, CROs and research sites.
04Industrial & laboratory automation
Robotic cells, inspection systems and instruments that have to hit uptime targets in an unglamorous environment.
05Training & simulation
Immersive training for clinical, industrial and field teams, with assessment models that prove competence transferred.
06Consumer health & wellness
Connected hardware and apps at the regulated edge, where a wellness claim and a medical claim are one design decision apart.
Selected work
Programmes,
not deliverables.
A sample of the kind of work we take on. Client names are withheld where our agreements require it.
View all workPoint-of-care screening with on-device inference
A vision model taken from a research notebook to a validated, offline-capable instrument, with the evaluation set and drift monitoring a submission actually needs.
Procedural training in VR with competence scoring
A headset-native simulator for a high-risk procedure, instrumented so that training managers get an assessment record rather than a completion tick.
Sample-handling robotics for a high-throughput lab
A mechatronic cell designed around the failure modes that actually stop a lab: mis-picks, consumable variance and the recovery routine nobody specifies.
Approach
Frame it, de-risk it,
then build it.
Almost every programme that goes badly went wrong before anyone wrote code. The wrong problem, an unvalidated assumption in the middle of the architecture, or a regulatory pathway chosen after the design was locked.
We spend the first weeks finding the parts of your programme most likely to kill it, and we test those first, while changing course is still cheap.
Why Sirotics
Built to carry a
programme, not a phase.
We are deliberately senior and deliberately small. You get the people who scoped your work doing your work, which is the whole reason the model holds together.
8
Practice areas that staff a single, joined-up programme team
Hardware
+ software Electronics, firmware, mechanical and cloud in one group, not two vendors
+ software Electronics, firmware, mechanical and cloud in one group, not two vendors
ISO 13485
aligned Design controls, risk and traceability run from day one, not bolted on
aligned Design controls, risk and traceability run from day one, not bolted on
2–4 wks
From first call to a scoped discovery sprint with a decision at the end
Senior by default
No pyramid. The engineers and designers in your kickoff are the ones on your programme in month nine.
Evidence-shaped from the start
Claims, endpoints and the evidence needed to support them are defined before the architecture hardens around them.
You own everything
IP, source, design history and toolchains are yours. We build so that your team can take it over, and we plan for that handover.
Insights
Notes from
the work.
What we're learning about shipping frontier technology into environments that were not designed to receive it.
All insightsYour model will change. Plan the change control now.
Predetermined change control plans are the difference between shipping a model update in weeks and re-opening a submission.
Usability engineering is not usability testing
Most teams discover the difference during a summative study, which is the most expensive possible moment to find out.
Decentralised trials fail on logistics, not technology
The platform is rarely the bottleneck. Device provisioning, connectivity and site burden usually are.
Sirotics is taking on new programmes
Tell us the part
you're worried about.
The fastest way to work out whether we're a fit is a 30-minute call about the riskiest thing on your roadmap. No deck required.