HealthPath — Healthcare Access for Underserved Communities

Inspiration

We are from three countries: Colombia, India, and Singapore. In all three, we saw the same story: people don't have trouble with the lack of medicine — people struggle to go through the process of bureaucracy in order to get it.

You are denied medication. You don't know your legal rights. You don't know where to look for it. You don't know how to file your complaint. So you just stop.

Our team thought: what if there is an AI that would go through the process instead of you?

What it does

HealthPath is an autonomous AI agent that takes your healthcare bureaucratic problem stated in plain language and provides:

Where you could actually buy the medication or resource
What kind of coverage you have by your insurance or scheme
What your rights are in the current situation
A ready-to-submit document (complaint, claim, request)

Hard boundary: It will never give medical advice. No diagnosis. No recommendations of treatment. If it detects an emergency, it will terminate and provide contact numbers of people who should be contacted — a human is responsible, not the system.

How we built it

Claude API — conversational understanding, document generation, multilingual support
n8n — autonomous workflow orchestration connecting resource databases and scheme registries
React/TypeScript — frontend with accessibility first
ElevenLabs — voice input/output for users with low digital literacy
Public health datasets + synthetic data — compliant with hackathon rules

The system works like this: you describe your situation → an emergency classifier runs first (if high-risk symptom language is detected, it stops and surfaces emergency numbers) → if safe, the agent infers the exact bureaucratic process you need → multi-step workflow executes in the background (querying APIs, checking eligibility, drafting documents) → you get a ready-to-submit formal document.

Challenges we ran into

The hardest part wasn't building the agent — it was defining where it stops. It's easy to say "no medical advice." It's harder to build a classifier that reliably catches symptom language before the system acts on it, especially across three languages and three different healthcare systems with different terminology.

We solved it by treating the emergency flag as a pre-response gate: if triggered, every other workflow is blocked and only emergency numbers surface. That way the system can't make a mistake in judgment — it defaults to "stop and ask a human."

Building across three time zones (Colombia UTC-5, India UTC+5:30, Singapore UTC+8) required disciplined async communication and a single source of truth for the architecture — n8n workflows became our shared spec.

Accomplishments that we're proud of

We shipped a real product in seven days across three countries with zero institutional backing
We built a classifier that doesn't need to be perfect — it just needs to be conservative. If there's doubt, it escalates to a human
Every feature we built serves the constraint: "no medical advice." That's not a limitation — it's what makes this legally defensible and actually useful. Healthcare navigation is a distinct problem from healthcare delivery

What we learned

Constraints become features. By refusing to diagnose, we carved out a problem space that's both legally defensible and genuinely useful
Agentic systems need hard exits. The most important design decision wasn't what the system does — it was where it stops and hands control to a human
Multilingual support at the agent level is non-negotiable. The system adapts to the user, not the other way around. Each locale demonstrates a real scenario: Colombia (EPS medication denial), India (Ayushman Bharat denial), Singapore (specialist wait time)
A three-person international team aligned on the problem first, then the architecture. Everything else followed cleanly

What's next for HealthPath — Healthcare Access for Underserved Communities

Pilot with community health workers in Colombia and India
Integration with government health scheme APIs (Ayushman Bharat in India, SISBÉN in Colombia)
Expand language support to Hindi, Spanish, and Tamil
SMS-based access for users without smartphones
Human case worker dashboard for tracking escalations and outcomes
Open-source the emergency classifier so other healthcare AI systems can use it safely