Now

Currently building

• Wrapping up the AdsRank platform re-architecture at Myntra — snapshot-based feature pipelines on Databricks/PySpark, with MCP agentic tooling for experiment automation across internal platforms.
• Async batch inference pipeline for multimodal LLM annotation at MIDAS Lab, using FastAPI + Celery for throughput-optimized processing.

Currently researching

• M.Tech thesis at IIIT Delhi — improving small LLM generation quality via Mixture of Refinement Agents. Exploring ensemble critique and targeted regeneration patterns.
• Learning-to-rank literature, specifically listwise approaches applicable to sparse ad auction settings.

Currently writing

• A post on idempotent, snapshot-based feature pipelines in Databricks — the pattern I wish existed before I had to build it.
• MCP-based agents for ML experiment automation: practical patterns for wiring Databricks, GitHub, and Grafana into a single agentic loop.
• Async batch inference for vision-LLMs with FastAPI + Celery.