Services
AI & Machine Learning
AI in healthcare is easy to demo and hard to ship. We build LLM features, models on device and clinical data, and the retrieval pipelines behind them — with the validation, monitoring, and audit trail that lets them reach production.
The problem
-
Leadership wants an AI story this year, and the proof-of-concept works on the demo laptop — but nobody can say how it behaves on real patients, real devices, or real PHI.
-
Your device generates terabytes of telemetry and waveform data that could predict failures or outcomes, and today it is archived and never read.
-
A general-purpose LLM vendor cannot tell you where your data goes, and your BAA, your security team, and your regulatory posture all need an answer.
How we work
- 01
Use-case & risk framing
We define what the model decides, who acts on it, and whether it falls under device software, clinical decision support, or operational tooling — before any training run.
- 02
Data pipeline
De-identification, feature extraction from device and clinical sources, and lineage you can reproduce — built on your data platform or one we stand up.
- 03
Model & integration
Fine-tuned or retrieval-augmented LLMs, classical ML on device telemetry, or edge inference on the device itself — integrated into the app, portal, or workflow where it is used.
- 04
Validate & monitor
Held-out validation, bias and drift monitoring, human-in-the-loop controls, and documentation designed to support the regulatory and security review that follows.
What you get
- A written AI use-case assessment: intended use, risk class, data sources, and the controls each one requires
- Reproducible data and training pipelines with lineage and de-identification
- Production model services or on-device inference integrated into your product
- Evaluation reports, drift and performance monitoring, and audit logging of every inference
- Documentation designed to support FDA AI/ML guidance, HIPAA, and your customers’ security reviews
Standards & technologies
- LLMs / RAG
- Claude / OpenAI / open models
- Python
- PyTorch
- ONNX / edge inference
- Vector search
- FHIR R4
- DICOM
- AWS / Azure
- HIPAA
Common questions
Can patient data stay inside our environment?
Yes. We deploy models and retrieval in your cloud account or on-prem, use vendor APIs only under a BAA with zero-retention terms, and run on-device inference where the data should never leave the device at all.
Is AI software a medical device?
Sometimes. The use-case framing step exists to answer that with your RA/QA team: a maintenance-prediction model is operational tooling; a model that informs a clinical decision may be software as a medical device, and we build accordingly.
Do you build on your own data, or ours?
Yours. Models that matter in healthcare are trained and evaluated on your devices, your patients, and your workflows. We bring the engineering, not a pre-baked model.
What about running models on the device itself?
Edge inference on embedded Linux is home ground for us — quantized models, ONNX or vendor runtimes, and the same signed-update path as the rest of the firmware.
Have data that should be doing more?
Bring us a device, an integration problem, or a product idea — we will come back with an approach, not a slide deck.