For my final project, I built MedTrust AI – Health Misinformation & Influencer Credibility Analyzer. MedTrust AI helps users verify public health-related videos before trusting or acting on them. The app analyzes a video URL, extracts the main health claim, checks trusted medical evidence, and generates a simple report with: ✅ MedTrust Score ✅ Risk Level ✅ Evidence Confidence ✅ AI Detected Claim ✅ Trusted Evidence Sources ✅ Red Flags & Missing Medical Context ✅ Safer Explanation and Recommended Next Step This project helped me learn how to build a complete AI application from idea to final demo, including Streamlit UI development, API integration, prompt engineering, evidence search, error handling, deployment, and responsible AI design. The biggest learning for me was that health-related AI tools must be built carefully, with safety, transparency, evidence, and disclaimers. 🔗 GitHub Repo: https://github.com/MLwithManali/MedTrust-AI 🔗 Live App: https://medtrust-ai-s4ede64magy2uvwq4rlynz.streamlit.app/ Thank you Decoding Data Science for this practical and valuable challenge!

Today I focused on final review and deployment for my project, MedTrust AI. I completed the GitHub repository, updated the README, added screenshots, tested the application, and deployed the working demo using Streamlit Community Cloud. MedTrust AI helps users verify health-related social media content by extracting health claims, comparing them with trusted medical evidence, and generating a clear misinformation risk report.

Today wasn't about adding new AI features. It was about making sure NovaDXB can survive real users. A demo only needs to work once. A product needs to handle bad inputs, unexpected behaviour, security risks, API limits, and edge cases you didn't think about when you started building. So today I focused on everything users never see: 🔒 Input validation and sanitization 🔒 Prompt-injection protection 🔒 Secret management and deployment security 🔒 Rate limiting and caching 🔒 Error handling without exposing internal details And, of course, bug fixing. One of the more interesting issues: cached responses were returning the AI answer but not the itinerary data needed for the side panel. The chatbot looked correct, but the experience wasn't. That's the kind of bug users notice immediately. The AI itself didn't get smarter today. The product got more reliable. With only 2 days left, NovaDXB is starting to feel less like a prototype and more like something people could actually use. 2 days to go. 🚀 Decoding Data Science Mohammad Arshad DDS Business Circle #BuildInPublic #AIAgents #AIChallenge #Dubai #BuildWithAI #GenerativeAI #Innovation #SoftwareEngineering #CyberSecurity #UX

Today I focused on final enhancements, security, and debugging for my project MedTrust AI. I improved input validation, prompt-injection safety checks, .env API key handling, JSON response validation, trusted medical source filtering, caching, latency tracking, and user-friendly error messages. I also tested edge cases such as invalid URLs, unsupported/private videos, API issues, and unclear health claims. Key learning: building an AI app is not only about getting outputs, it is also about making the app safe, reliable, and user-friendly. MedTrust AI: Helping you verify health information before you trust it.

Today I worked on integrating my AI application backend with the user interface. I focused on API connection, input validation, error handling, and end-to-end testing to make the workflow smoother and more user-friendly. Key learning: a good AI app needs both strong backend logic and a simple, reliable interface.