Intake

Intake was inspired by a simple frustration: health data is everywhere, but understanding it is still too hard. Lab results, clinician notes, medication lists, symptoms, and follow-up tasks often live in separate places, leaving patients to connect the dots alone.

I built Intake to turn scattered records into a patient-owned health graph. Users can import mock EHR data through Healthex, talk through their history by voice, add clinician notes, explore a knowledge graph, ask questions with cited answers, and generate visit-ready reports.

The project is built with Next.js, Supabase, xAI Grok, streaming agents, and a graph-based data model. The agents use tools to search patient evidence, compute trends, inspect relationships, and return cited outputs instead of unsupported summaries.

I learned a lot about making AI feel trustworthy. The hardest parts were streaming tool calls smoothly to the UI, keeping citations aligned with graph nodes, handling model failures when it answered in prose instead of structured tool calls, and designing fallback paths that never discard useful work.

At its core, Intake is about reducing the gap between having health data and understanding it:

https://github.com/krishh-p/intake

Built With

• cursor
• grok
• next.js
• vercel