Muhammad Taha
Software Engineer building AI-powered products with LLMs
I built Aegis, a Flood Prediction and Disaster Assistance System. It forecasts floods in Pakistan's river systems and helps people as water rises.
It's a PWA, so it works on a phone with nothing to install.
The goal was simple: when someone is scared and the water is coming up, the web app should show them what matters, fast. That's why the whole frontend is built around a map, and why sign-in is kept short.
For the forecasting, I trained GRU and ED-LSTM models, and the backend runs on FastAPI with Firebase.
There's also AegisAssist, a bilingual chatbot. Answers stream in as they're written, and a RAG pipeline makes it answer from real sources instead of guessing.
What do you think? 👇
0
15
Architected and developed an AI tutoring platform that encourages conceptual learning through collaborative multi-agent workflows instead of answer-focused interactions. Designed a four-agent architecture (Supervisor, Teacher, Critic, and Librarian) with adaptive Socratic teaching and asynchronous coordination using Redis and Celery. Implemented asynchronous task processing to improve scalability, maintainability, and separation of responsibilities across AI agents.
1
56
Engineered an AI legal reasoning framework to improve the reliability of LLM-generated legal responses through evidence-grounded reasoning. Designed an evidence-verification pipeline using a causal cross-attention gating mechanism to validate generated responses against retrieved legal source material. Reduced unsupported AI-generated legal conclusions by enforcing evidence-grounded reasoning before presenting legal answers.
0
58
First Aid Learning
0
0