I've built AI systems at the National Center of Artificial Intelligence, including a voice-to-report pipeline using Whisper STT and MedGemma that reduced report generation time by 60%.
I led work on real-time clinical reporting, building WebSocket ETL pipelines, speaker diarization across 50+ recordings, and a FHIR-compliant MongoDB data layer supporting over 1,000 patient profiles. I also deployed quantized MedGemma models on GPU clusters, delivering 3x faster inference latency.
At the Pakistan Telecommunication Authority, I designed Node.js and Express REST APIs and dashboards for administrative SQL data access across MySQL and PostgreSQL.
My projects span production ML observability, brain MRI synthesis with WGAN-GP, dementia diagnostics with ResNet-34 and RAG, and CUDA optimization that achieved a 56x speedup on an MNIST neural network.
