At Superside, I architected and built a multi-agent AI Project Manager with 100+ tools, reusable skills, and human-in-the-loop escalation. It runs project workflows end-to-end and reduced manual effort across core PM workflows by 50%+.
I also built RAG-powered assistants for structured data and internal knowledge, plus Pulse, a real-time AI signal detection system processing about 2.2K customer conversations per day. I added production-readiness controls for agents, including versioned prompts and skills, Langfuse tracing, and golden-set evaluation.
At Careem, I owned the roadmap and technical direction for core ML Platform components, including a governed Feature Store supporting 5K+ features across 15+ teams. I scaled online serving, led a 100TB+ zero-downtime TiDB-to-TiKV migration that cut storage and compute costs by about 80%, and re-architected Apache Pinot workloads to reduce infrastructure costs by about 60%.
