Healthcare AI assistants that show where each answer comes from.

We build AI features for clinics and health products: assistants that answer from your documents and records, summaries and drafts for staff. Every answer cites its sources, and each clinician sees only their own patients.

Phone screens with a PHQ depression question and a stress level result

What does healthcare AI development involve?

Onix Health builds AI features for clinics and health products: assistants that answer from your documents and records and cite the source of each claim, plus draft notes, letters and summaries that a clinician reviews and signs. We run Claude through AWS Bedrock or Anthropic directly, and the system can switch providers. Each clinician sees only their own patients, and every change is tested against a set of real examples with known answers.

What we build: healthcare ai

1

Assistants that cite sources

Answers built from your documents, guidelines or records, with a link to the source of each claim so staff can check it.

2

Summaries and drafts

Draft notes, letters and patient summaries that a clinician reviews and signs.

3

Search across your records

Plain-language search across documents and data, limited to what each user is allowed to see.

4

Claude on AWS Bedrock or Anthropic

We run Claude through AWS Bedrock or directly through Anthropic, and the system can switch providers.

5

Testing before launch

A set of real examples with known answers, run on every change, so you can see if quality drops.

6

Access control

Each clinician sees only their own patients, and the AI only reads what that user may read.

How the project runs

  1. 1

    Free technical audit

    Five working days. We look at your current system, data and plans, and you get a written report with what we found and what we would do.

  2. 2

    Pick one task and a test set

    We choose one job for the AI, collect real examples with the right answers, and agree how we will measure it.

  3. 3

    BAA before patient data

    We sign a BAA with you and run models through providers that sign one too, such as AWS Bedrock.

  4. 4

    Build, test, launch

    We test the assistant against the example set, let clinicians review its answers, and pass a penetration test before launch.

Technology and standards

  • Claude
  • AWS Bedrock
  • Anthropic API
  • Python
  • Node.js
  • TypeScript
  • React
  • PostgreSQL
  • AWS
  • FHIR R4
  • Built to the HIPAA Security Rule
  • BAA signed before any patient data is shared
  • AI answers cite their sources
  • Penetration test passed before launch
  • Each clinician sees only their own patients

Questions about healthcare ai

1

Can AI be used with patient data under HIPAA?

Yes, if every vendor that handles the data signs a BAA and access is limited. We sign a BAA with you and run models through providers that offer one, such as AWS Bedrock.

2

Which AI models do you use?

We use Claude through AWS Bedrock and directly through Anthropic. The system can switch providers, so you are not tied to one.

3

How do you stop the AI from making things up?

The assistant answers from your documents and records and cites the source of each answer. Staff can check every claim, and we test answers against a set of real examples before launch.

4

Will the AI show one clinician another clinician's patients?

No. Each clinician sees only their own patients, and the AI reads only what that user is allowed to see.

5

Does the AI make clinical decisions?

No, our assistants support staff and a clinician makes the decision. If you want a feature that works closer to a diagnosis, we plan it with your regulatory team first.

6

How do we start a healthcare AI project?

Start with the free five-day audit. We look at your data and the task, and you get a written report on what an AI feature could do and how we would test it.