At Irvinei, I designed a multi-agent voice-control system for IoT devices using LangGraph, LangChain, and FastAPI. I also trained BERT-based classifiers that routed queries to the right agent with 95% accuracy.
I built a RAG-based product-support agent that reached 87% response accuracy with 3–4 seconds average inference latency. By optimizing the voice agent, I reduced response time from 15 seconds to under 4 seconds.
I also trained a YOLO detector for an AI home security system on a 100K-image dataset, improving mAP from 0.58 to 0.87 at 90% accuracy. I handled the ML lifecycle from data collection and labeling through deployment with backend and DevOps teams.
My projects include TrainCortex, an autonomous ML pipeline, and SentiSense, a DistilBERT model for sentiment classification in English, Urdu, and Roman Urdu. Earlier, as a Software Engineering Trainee at KoderLabs, I developed REST APIs with ASP.NET Core and Entity Framework.
