AUDIO → TEXT
Session transcription
Speech models fine-tuned for clinical vocabulary, medication names, and the cadence of a therapy session. Real-time on the Neural Engine.
On-device clinical AI · Canada
The ambient scribe that never sends your sessions anywhere.
Transcription and structured notes, end-to-end on the iPhone in your pocket. No cloud step, no upload, no BAA gymnastics. Built for therapy, psychiatry, and privacy-sensitive Canadian practice.
What we do
01 · Capabilities
TRANSCRIPT → NOTE
Structured notes, on-device
Small language models that turn a conversation into a SOAP or DAP note you'll actually sign — without the transcript ever leaving the device.
IMAGE → TEXT
Paperwork and intake capture
Photograph an intake form, medication label, or handwritten note and get clean structured text back. On-device vision, same privacy posture.
How this is different
02 · The wedge
End-to-end on the device
Most "on-device" scribes transcribe locally and then upload the transcript for note generation. We run both steps on the phone. The session audio, the transcript, and the draft note never leave the Neural Engine.
Built for therapy first
General clinical scribes miss the things therapists actually care about — affect, tone, risk language. Our models are fine-tuned on therapy-shaped conversations, not annual physicals.
Canadian data residency, by architecture
PHIPA, PIPEDA, BC PIPA — the on-device posture maps cleanly to every provincial framework. If the data never moves, residency becomes a non-question.
No BAA needed for the core product
We can't see your session data, so we don't need to sign for the right to protect it. That alone cuts weeks off most procurement cycles.
You own the weights
Tier 3 engagements deliver a fine-tune of our models on your practice's vocabulary and workflow. It's a deliverable — not a dependency on someone else's roadmap.
Predictable cost
A flat license per clinician, not a bill per token. Usage scales with your practice, not your API invoice.
How we work
03 · Approach
-
STEP 01
Scope
Two-week discovery: the workflow, the data, the device, the evaluation that would actually convince a clinician. We write it down.
-
STEP 02
Fine-tune
Data curation, supervised fine-tuning, and domain evals that measure what matters in a session: vocabulary accuracy, hallucination rate, safety on risk-language moments.
-
STEP 03
Ship
Core ML / MLX / ONNX builds, a thin SDK, and on-device latency benchmarks on your actual hardware. You own the weights and the integration.
Team
04 · Who you work with
photo
Bis Ghosh
CTO
Training, inference, and shipping models to edge devices. Owns the technical roadmap.
photo
Teja
Context Engineer
Turns clinical workflows into the prompts, data, and evals that make a fine-tuned model actually useful in a visit.
photo
Susmita
Go-to-Market
Partnerships with clinics and clinical-tool vendors. Owns pilots, pricing, and how we reach the practices that need this.
Sessions that shouldn't leave the room.
We're running a small closed pilot with therapists and small Canadian practices. A 30-minute call tells us whether your workflow fits.
bis@neuralrust.network