Skip to content
Mirun Kaushik
← All work

02Hackathon · 24 hours20253rd place, KRCE hackathon 2025

Health Link

Turns scattered medical reports into a graph a patient can understand

My part
Hackathon team
Status
Built
Stack
Tesseract OCR, Neo4j, Hugging Face
Code
GitHub ↗on a teammate’s account

Fig. — How data moves through it

  1. 01Scans & PDFs
  2. 02Tesseract OCRtext extraction
  3. 03Ontology rulesstructure, no LLM
  4. 04Neo4j graphreports as relationships
  5. 05Medical LLMjargon + body mapping(model call)
  6. 06Chat · 3D body · lessons
Model call Rules, data, queues

01Problem

Medical records sit on paper or in scattered files, with no good place to keep them or make sense of them.

02Approach

OCR the reports, then use ontology rules to structure the text into a Neo4j graph, so a model isn’t needed for every step. Models come in only where rules can’t help: explaining jargon, and mapping findings to the body.

03What I built

  1. 01The ingestion pipeline: upload scans or PDFs, Tesseract extracts the text, ontology rules structure it, and it is stored as graph relationships in Neo4j.
  2. 02A chatbot over the graph that explains medical jargon in the patient’s own reports.
  3. 03An interactive 3D anatomy model: a medical LLM on Hugging Face maps each report to the affected body parts, which light up on a spinning skeleton. Clicking a part shows its findings.
  4. 04A Duolingo-style learning flow (my idea): each uploaded report unlocks a chapter with a 7-card lesson and 2–3 quiz questions, with streaks.

04Built by my teammates

  • The original product idea.
  • Smartwatch integration, tested with a Google watch and a Fitbit.
  • The ML model that combines watch metrics with medical history.