Project Story

Inspiration

The inspiration for Health Info Analyzer came from seeing how many people trust short health videos on social media without verifying the information. Platforms like Instagram, TikTok, and YouTube Shorts are flooded with advice that is often misleading or risky. I wanted to create a simple tool that helps people quickly check whether the health advice they see is Safe ✅, Uncertain ⚠️, or Risky ❌.

What I Learned

During this project, I learned:

How to extract audio from video URLs using yt-dlp.
How to transcribe audio to text with OpenAI's Whisper model.
How to call Hugging Face GPT-OSS APIs to analyze text.
How to store and manage analysis history securely on a device.
Basics of backend development in Go and frontend in React Native (Expo).

I also improved my problem-solving and debugging skills, especially when handling unexpected API responses and audio transcription errors.

How I Built It

Backend:
- Built in Go, it handles video analysis requests, calls Whisper for transcription, and GPT-OSS for analysis.
- Stores results locally in SQLite.
Frontend:
- Built in React Native (Expo).
- Users can paste video links, see results, and review their history.
Integration with AI:
- Used Hugging Face GPT-OSS API to classify health advice as Safe, Risky, or Uncertain.
- Stored API keys securely in .env files.

Challenges Faced

Handling different video formats and extracting clean audio from Instagram Reels.
Dealing with Whisper transcription errors for noisy audio.
Managing unexpected or inconsistent responses from the GPT-OSS API.
Ensuring user privacy, so all data is stored only on the device.