At CabsWale, I built the AI stack from zero, including Suvidha, an agentic browser-automation platform that handles 10,000+ tasks a day. It has completed 1M+ tasks on a self-managed GCP VM fleet, cutting handling time to under 2.5 minutes per task.
I also built Raahi, a real-time Hindi/Hinglish voice assistant for driver calls, and paired automation with fine-tuned LLMs to bring driver ticket resolution down from 1–2 days to 10–15 minutes. I lead DevOps and MLOps for the platform, which runs on GCP and GKE.
Outside CabsWale, I’ve contributed 11 merged PRs to OpenTelemetry Collector. My projects include a PagedAttention LLM serving engine and GPT training and inference work in Python and Rust.
