From scattered health records
to a source-linked Portrait.
Surforme. A consumer health application built end to end in approximately three to four months. Its responsive web experience brings uploaded records, personal context and health signals into a Portrait users can explore over time.
- Context
- Surforme. An independent consumer AI product.
- Service
- End-to-end product development, AI pipelines, source traceability, automated evaluation and cloud infrastructure.
- Industry
- Consumer health, health intelligence
- Specialty
- Build and integrate AI
- Status
- Live beta. Access activated by the team.
Overview
Surforme brings health records, personal context and connected wearable data into one place. Source-linked Portraits help users prepare for conversations, identify gaps to investigate and explore changes across successive reports.
The responsive web application is in beta, with access activated by the team. Approximately 100 accounts have been registered and activated, with around 40 users active weekly and nearly all returning over a month.
Challenge
Health information accumulates across documents, devices and appointments. Bringing it together requires more than generating a summary: extracted values need validation, documents need to belong to the right person, and findings need traceable support.
Solution
A five-stage pipeline parses documents, extracts and validates signals, generates a Portrait, evaluates it against the evidence and delivers approved results.
Identity checks hold mismatched documents out of synthesis until ownership is confirmed. References are resolved against known sources; unsupported priority items are removed, and insights without supporting signals are labelled as hypotheses.
Before delivery, each Portrait must pass seven automated checks combining deterministic rules and a separate model evaluation. A failed check holds the new Portrait and preserves the previously delivered version. Delivery is automated, with no human review step.
The infrastructure combines Canadian storage and Montréal application compute with consent-gated processing by external AI providers. Documents can include identifying information when sent for processing. Background jobs carry record IDs rather than health information in their payloads.
Results
Built in months.
Used throughout the beta.
- Delivery time
- Approximately 3–4 months to build the product end to end
- Activated accounts
- Approximately 100 registered and activated beta accounts
- Weekly activity
- Around 40 users active weekly
- Monthly activity
- Nearly all 100 activated users returning over a month
- Depth of use
- Some users have processed more than ten health records and generated multiple Portraits
- Delivery controls
- Seven automated checks must pass before a Portrait is delivered
Build time and recent activity figures are owner-reported estimates as of September 2026. Activity reflects returning and using the product; these figures are not a cohort retention analysis. Depth-of-use examples describe some users. Early feedback is qualitative, separate from measured health outcomes or a quantified accuracy evaluation.
- Early user feedback highlights preparation, useful gaps to investigate and less repetitive manual input.
- Successive Portraits help users explore changes in their information over time.
- Failed quality checks hold delivery and preserve the previously delivered Portrait.
Relevant experience for end-to-end AI product delivery, source traceability, automated evaluation and cloud infrastructure.