Software Development Engineer 2 · Wadhwani AI
Jan 2024 — Present- Built a standalone spoken English assessment platform where ML models score reading, speaking, vocabulary and listening tasks, with adaptive practice pathways. 100,000 students onboarded, 2.5M assessments served, scaling to 500,000.
- Ran the serving setup for compute-heavy inference windows: 100,000+ assessments at 25 concurrent ML inferences per minute (40s average latency) with no downtime, after reworking queue management and resource allocation.
- Shipped LLM features in production, including Llama-based paragraph generation and a “Lecture Segmentation” feature built in two weeks.
- Designed the search engine for a MOOC platform on BigQuery with custom ranking. 500,000+ searches served, 180,000+ course clicks driven.
- Built a configurable field data collection platform with offline and online inference pipelines, audio/camera/location capture, and a data-quality portal. 5,000 collectors and 200,000 entries in four months.
- Deployed a model on an on-premise server inside a hospital PACS workflow, and built offline-first radiology and dermatology apps handling bidirectional sync, image caching and device memory limits.
- Wrote a cloud-agnostic object storage wrapper now used by 5 projects across the organisation.