• 

  Carelink Main

• 

  Carelink Gemini Brian

• 

• 

  CareLink Logo

What it does

CareLink is a voice-first medical simulator and intelligent triage agent. Designed with a high-contrast, "frictionless" interface that mimics the familiar layout of WhatsApp, it allows users to simply speak their symptoms in their native or mixed languages (such as "Tanglish").

Intelligent Triage: Powered by Google Gemini, it listens to voice notes, analyzes clinical urgency, and detects life-threatening emergencies (like heart attacks) in real-time.

Live Dashboard: It captures precise geolocation and synchronizes patient data instantly to a Doctor Dashboard, allowing medical professionals to view AI-generated clinical summaries and accept cases immediately.

How I built it I built the application as a web-based platform with a focus on speed and accessibility using Google AI Studio.

The Intelligence: I utilized Google Gemini to process unstructured voice inputs. Its ability to parse code-mixed language allowed me to extract accurate medical data from casual speech.

The Backend: I architected the system using Supabase Realtime. This ensures that as soon as a patient speaks, the data appears on the doctor's screen with sub-second latency.

The UI: I designed the frontend to look exactly like WhatsApp because it is an interface my target users already know, removing the learning curve entirely.

Challenges:

Language Parsing: Ensuring the AI could accurately interpret "Tanglish" (Tamil-English mixing) without losing critical medical context was difficult.

UI Discipline: My biggest struggle was stripping the interface down. As a developer, I wanted to add features, but I had to constantly remind myself that for this demographic, every extra button is a barrier.

Latency: Optimizing the voice-to-text-to-dashboard pipeline to ensure it felt like a real-time conversation rather than a form submission.

Accomplishments that I am proud of I am most proud of the "Zero-Training" nature of the app. Seeing the system successfully take a vague, mixed-language voice note and convert it into a structured, medically accurate summary on the dashboard felt like a breakthrough. I'm also proud of the production-ready architecture; the Supabase integration allows the app to handle real-world scale right out of the box.

What I learned I learned that the best User Interface for non tech savvy users is often no interface at all—just voice. I discovered that Large Language Models are not just for generating text. they are powerful accessibility tools that can bridge the digital literacy gap, turning natural language into a reliable command line for complex systems.

What's next The architecture is designed to be extensible. Next, I plan to:

Expand the Language Library: Ingest more regional language models to support users across different states and dialects.

Wearable Integration: Connect CareLink to smartwatches for passive health monitoring (heart rate/fall detection) that triggers the voice agent automatically.

Built With

• api
• gemini
• gemini3
• gemini?s
• google
• model
• native
• natural
• natural-language-processing
• react
• supabase
• tailwindcss
• text-to-speech
• tts)
• typescript