Ammar Jawed

I operate at the intersection of research depth and production engineering, designing and implementing models “from first principles” (custom loss functions, novel architectures, rigorous implementations) and then owning the full path to deployment. In my current role, I led end-to-end architecture and production delivery of AI pipelines that convert complex documents into structured relational JSON, architected fault-tolerant asynchronous backends with retry guarantees and audit logging, and significantly reduced manual post-processing overhead at scale.

I’ve also delivered a full-stack sales intelligence platform with agentic workflows and spatial modelling, including intelligent data dashboards, CRM-style activity management, authentication, and user administration—backed by FastAPI and deployed on Azure via Docker with a complete stack migration completed without disruption. Alongside production work, I lead applied research (spatial flow modelling and rigorous model-vs-baseline evaluation) and ship systems like stateful LangGraph-driven analytics, adaptive OCR pipelines, and distributed PyTorch training on 8× A100 GPUs.