labs@atheron:~/industries/healthcare/ai$ train agent --grounded

AI for healthcare

Clinical notes and referral intake, on private models.

In healthcare the best AI work is the paperwork around care: the note after the visit and the referral before it. Both touch health records, so we build them on models that run where you decide, and a clinician or staff member always has the final word.

Start a project See the example projects

// what matters here

What is different in this industry.

Records stay with you

Open-source and custom models run on your own GPU server or in your own cloud account in Canada. No visit audio or chart text is sent to an outside AI provider.

The clinician signs, not the model

The assistant prepares a note; the clinician reviews, edits and signs it. Nothing is filed to the chart without them.

Measured on your own cases

Each model is scored against a held-out set of your de-identified notes or referrals before it goes live, and again before every change.

Paperwork, not diagnosis

We build tools that handle documents and routing. Software that makes clinical decisions can fall under Health Canada medical device rules, and we scope that with your regulatory team.

// example projects

Priced examples for this service.

Each one is hypothetical, labelled as such, and priced live from our rate card. Open any of them in the estimator and make it yours.

Example projectWeb appAIOpen-source model

Private clinical note assistant

Hypothetical. Not a client, not a result.

// The problem

Clinicians spend their evenings finishing notes, and the clinic cannot send recorded visits to an outside AI service.

// What we would build

An assistant that transcribes the visit with the patient's consent and prepares a structured note for the clinician to review, edit and sign, on open-source models running on a GPU server in the organisation's own data centre, so no audio or text leaves it.

// What is in it

Consent captured before recording, and recording that stops on request
Speech to text and a structured note in the clinic's own templates
The clinician reviews, edits and signs; nothing is filed without them
Notes filed to the EMR through its interface
An evaluation set from real, de-identified visits, run before every change

// Stack

Whisper (open-source)
Llama or Qwen (open-source)
vLLM
Next.js
Lenovo GPU server on site

// estimate

Build: ≈ US$155,000 to US$236,000, delivered within 39 weeksCAD 221,000 to 336,500
Hosting: ≈ US$176 a monthCAD 250 a month
Support: ≈ US$2,500 a monthCAD 3,565 a month
Hardware: ≈ US$47,800 to US$197,000CAD 68,000 to 280,800

Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.

AI route: Open-source model
Model running cost: ≈ US$176 a monthCAD 250 a month

Timeline by milestone

Discovery: 2.9 to 3.4 weeks
Specification and evaluation plan: 0.4 weeks
Design approved: 2.4 to 4.4 weeks
Core features: 1.4 to 2.4 weeks
Full build: 0.9 to 1.4 weeks
Working pilot: 4.9 to 8.4 weeks
Testing and fixes: 0.4 to 0.9 weeks
Controls built and evidenced: 0.9 to 1.4 weeks
Readiness review: 2.9 to 14.4 weeks
Launch: 0.9 to 1.4 weeks
Production: 0.9 to 1.4 weeks

Outside our hands, and added to the calendar

Hardware delivery, after it is ordered: 2 to 6 weeks
EHR vendor approval and test access: 1 to 4 weeks
Business associate agreements with your vendors (HIPAA): 2 to 6 weeks
Booking a penetration tester and their report: 2 to 4 weeks

How it is paid

Deposit 20%: CAD 44,140 to 67,240
Discovery 1.5%: CAD 3,310.50 to 5,043.00
Specification and evaluation plan 8%: CAD 17,656 to 26,896
Design approved 1.5%: CAD 3,310.50 to 5,043.00
Core features 4.6%: CAD 10,152.20 to 15,465.20
Full build 3.1%: CAD 6,841.70 to 10,422.20
Working pilot 24.2%: CAD 53,409.40 to 81,360.40
Testing and fixes 1.5%: CAD 3,310.50 to 5,043.00
Controls built and evidenced 7.7%: CAD 16,993.90 to 25,887.40
Readiness review 7.7%: CAD 16,993.90 to 25,887.40
Launch 1.5%: CAD 3,310.50 to 5,043.00
Production 8.7%: CAD 19,200.90 to 29,249.40
Holdback, 30 days after launch (10%): CAD 22,070 to 33,620
GPU time for training and testing, at cost: CAD 300. Billed up front, at cost, outside the milestones
GPU server for AI, for the model: CAD 68,000 to 280,800. Billed up front, at cost, outside the milestones

Start from this example

Example projectWeb appAICustom LLM

Referral intake and routing model

Hypothetical. Not a client, not a result.

// The problem

A specialist clinic receives referrals as faxes and scans in dozens of layouts, and staff read every page to find the reason, the urgency and what is missing.

// What we would build

A custom model fine-tuned on the clinic's own de-identified referrals that reads each one, pulls out the fields, flags missing results and suggests a queue, with a person confirming every referral before it is booked.

// What is in it

Faxes and scans read into structured fields: reason, urgency, referring clinician
Missing test results flagged, with a request to the referrer prepared
A suggested queue and priority, confirmed by staff
A model fine-tuned on your referrals, scored against a held-out set
Served in your own cloud account in a Canadian region

// Stack

Fine-tuned open-weight model
OCR pipeline
Python
Next.js
AWS (Canada region)

// estimate

Build: ≈ US$177,000 to US$270,000, delivered within 33 weeksCAD 251,700 to 384,300
Hosting: ≈ US$6,710 a monthCAD 9,550 a month
Support: ≈ US$2,500 a monthCAD 3,565 a month

Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.

AI route: Custom LLM
Model running cost: ≈ US$6,530 a monthCAD 9,300 a month

Timeline by milestone

Discovery: 2.9 to 3.4 weeks
Data and evaluation plan: 0.4 weeks
Design approved: 2.4 to 3.9 weeks
Core features: 1.9 to 3.4 weeks
Full build: 1.4 to 2.4 weeks
Training: 7.4 to 12.4 weeks
Testing and fixes: 0.4 weeks
Evaluation and red-teaming: 3.9 to 4.4 weeks
Launch: 0.9 to 1.4 weeks
Production: 0.9 to 1.4 weeks

Outside our hands, and added to the calendar

EHR vendor approval and test access: 1 to 4 weeks

How it is paid

Deposit 20%: CAD 50,100 to 76,620
Discovery 1.5%: CAD 3,757.50 to 5,746.50
Data and evaluation plan 7.9%: CAD 19,789.50 to 30,264.90
Design approved 1.5%: CAD 3,757.50 to 5,746.50
Core features 4.6%: CAD 11,523.00 to 17,622.60
Full build 3.1%: CAD 7,765.50 to 11,876.10
Training 23.9%: CAD 59,869.50 to 91,560.90
Testing and fixes 1.5%: CAD 3,757.50 to 5,746.50
Evaluation and red-teaming 15.9%: CAD 39,829.50 to 60,912.90
Launch 1.5%: CAD 3,757.50 to 5,746.50
Production 8.6%: CAD 21,543.00 to 32,946.60
Holdback, 30 days after launch (10%): CAD 25,050 to 38,310
GPU time for training and testing, at cost: CAD 1,200. Billed up front, at cost, outside the milestones

Start from this example

// questions

Questions we get about this.

Why not use a frontier model for clinical notes?

You can, where your privacy officer approves the provider and the agreement. Many organisations prefer an open-source model on their own server so recorded visits never leave; we price both.

How do patients consent to an AI note assistant?

Consent is asked and recorded before each recording, and the patient or clinician can stop it at any moment. Your privacy team sets the wording.

What does a referral model need to learn from?

A set of your past referrals, de-identified, with the fields staff pulled out of them. We measure as we label and tell you when more examples would help.

Back to the industry: Healthcare About the service: AI agents and models

// next

Not quite your project?

Tell us what you have in mind. We will come back to you with a range and the questions that would narrow it. Or book a call and talk it through.

Start a project Book a call