Industries
AI for healthcare that knows when to call a human.
Digital health teams come to us with the same tension: users want answers now, and safety demands a professional in the loop. We build both sides. Patient-facing AI that advises without diagnosing, and the escalation paths, review queues and audit trails that keep clinicians in charge. Working worldwide since 2019.
What does AI for healthcare look like in production?
Advisory systems, not diagnostic ones. In production that means patient-facing products that assess and guide: a skin assessment from a selfie in under a minute, camera-based physiotherapy that reads movement form, conversational health support on WhatsApp, and care operations like matching care workers to clients. Every system we build states its advisory limits, escalates to a professional on defined triggers, and runs on a security posture designed to support HIPAA, GDPR and DPDP expectations.
The problems we see in healthcare
Healthcare carries the highest cost of being wrong and a rising cost of doing nothing. The teams who call us, funded digital health startups and care operators, are stuck between the two. The patterns repeat.
- Users expect instant answers; clinicians cannot be on call for every question.
- Engagement dies between visits: exercises skipped, plans abandoned, no signal back to the care team.
- Intake, triage and matching run on spreadsheets and phone calls while demand grows.
- Back-office work eats clinical hours: notes, reports, scheduling, follow-ups.
- AI pilots stall in privacy review because nobody designed for it from day one.
- Chatbots that overpromise. In health, a confident wrong answer is a risk to a person, not just a bad review.
What we build for healthcare teams
Advisory systems with hard boundaries. Everything we ship in health states what it is (guidance, not diagnosis), knows what it may not answer, and hands off to a professional when the conversation crosses the line. Within those boundaries, the range is wide.
- Patient-facing advisory products: assessment from a selfie or a camera, guidance in plain language, on web, mobile or WhatsApp. Our AI product development practice, entered through a scoped Prototype Sprint.
- Conversational health support that listens, guides and escalates on defined triggers. Agentic AI development, starting with a scoped Agent Readiness Audit.
- Clinical guardrails: review queues, escalation rules and accuracy dashboards run by trained reviewers. Human-in-the-loop operations, starting with a scoped Accuracy Gap Assessment.
- Care operations: intake, matching, scheduling, reporting and documentation automated with audit trails. AI process automation, one workflow fast.
- A roadmap scored against your data, your compliance exposure and your clinical reality. AI consulting, a scoped AI Opportunity Map.
Proof from production
Health work we can show, and health work we can only describe. The NDA projects stay unnamed; the engineering carries over.
- Skin analysis and advisory: a consumer health product that turns a selfie into a skin assessment and personalized guidance in under a minute.
- Movement intelligence: Torq Labs smart wear plus Fisio, camera-based physiotherapy that reads movement form and injury risk live.
- Under NDA: a WhatsApp-native digital therapist, built advisory-first with defined escalation to human professionals.
- Under NDA: care-worker matching built on vector search, pairing care needs with the right worker.
How an engagement starts
A 30-minute founder call, then a scoped entry engagement. Small and honest: if the answer is that you should not build yet, that is the report.
- Prototype Sprint: your digital health product bet as a working system on real data, with eval numbers.
- Accuracy Gap Assessment: where your AI fails, what each failure costs, and the human layer that closes the gap.
- Agent Readiness Audit: one care workflow mapped and scored for agent-fit, with a go or no-go.
- AI Opportunity Map: readiness, scored use cases, and an execution roadmap grounded in reality.
What a similar project needs
Before the first model call, a health project needs boundaries and plumbing. We have built this enough times to hand you the checklist.
- A written advisory boundary: what the system may say, what it must not, and who signs it off.
- Escalation paths designed with professionals, not appended after launch.
- Consent and data handling mapped before build: PII masking, least-privilege access, data residency, audit logs.
- A security posture designed to support HIPAA, GDPR, DPDP and SOC 2 expectations when the audits come.
- Evals that score safety, not just accuracy: does it refuse what it should, does it escalate when it must.
- A human review loop sized to the risk, shrinking only as accuracy earns it.
Do you build diagnostic AI?
No. We build advisory systems. They assess, guide and flag; they do not diagnose, prescribe or replace a clinician. Every health system we ship states that boundary to the user and escalates to a professional on defined triggers. The line is a design decision, made with you and your clinical advisors on day one, not a disclaimer added before launch.
How do you handle health data and HIPAA?
We design for it rather than claim it. Least-privilege access, PII masking before models see data, audit logs on every action, data residency honored, and human approval gates on sensitive steps. That posture is designed to support HIPAA, GDPR, DPDP and SOC 2 expectations. Your auditors and counsel make the compliance call; we build so that call is easy.
Can AI talk to patients safely?
Within boundaries, yes. The conversational systems we build follow rubrics reviewed by professionals, refuse what they should not answer, and hand off to a human when the conversation crosses defined lines: risk language, clinical questions, signs of distress. A WhatsApp-native digital therapist we built under NDA runs exactly this way, advisory-first with human escalation.
We are a digital health startup, pre-launch. Where do we start?
Usually a Prototype Sprint: your product bet as a working system on real data. You get the prototype, the eval numbers, and an honest read on what production takes, including the advisory boundary, the escalation design and the security work. Book a founder call and bring the idea.
Our health AI pilot stalled in privacy or clinical review. Can you rescue it?
That is a common starting point. Health pilots usually stall on the missing layer: evals, guardrails, escalation paths, audit trails, and a data-handling story that survives review. We audit what exists, add that layer, and either productionize the pilot or tell you plainly why it will not pass.
Talk to the people who build.
One call. An honest read on what AI can do for this, and the number it has to beat.