I work across the ML and infra stack, pipelines, deployment, and MLOps. Most of what I know I picked up building real products end to end rather than from coursework alone, so I'm comfortable owning a problem from raw data through to something running in production.
Meru — geospatial data platform, India
Worked across the technical stack: ML pipelines, deployment, MLOps, and the geospatial data infra underneath it all. Handled data source integration for satellite and atmospheric layers, plus architecture decisions for how vector and NetCDF data get stored and queried at scale.
Python DuckDB GeoParquet Zarr Satellite EO data
VoClyp — conversation intelligence for field sales
A B2B platform that records in-person sales visits, transcribes across 120+ Indian languages, and routes structured insights to reps, agents, and leadership. Contributed across product, GTM materials, and outreach.
Speech-to-text NLP Multilingual pipelines
Plateau — semantic circuit breaker for autonomous agents
Built this for the Frontier 2026 hackathon, in the AI Safety and Observability track. It detects when an autonomous AI agent is stalling by tracking action-similarity and observation-novelty signals together, and steps in before the agent burns through its budget looping. Delivered with full documentation, a pitch deck, and a business case.
Agent observability AI safety
Flood segmentation with Prithvi EO — Kaggle, AISEHack
Built a semantic segmentation pipeline on IBM's Prithvi EO V2 300M foundation model with a UperNet decoder. Reached 0.704 mIoU and finished top-50. Most of the real work was debugging: NumPy and SciPy version conflicts, GPU OOM issues, and normalization mismatches across satellite data sources.
PyTorch TerraTorch Remote sensing Foundation models
Core
Geospatial / EO
Infra
Open to internships and full-time roles, and always happy to talk about ML, geospatial data, or anything running in production.
VIT Chennai · IEEE Vehicular Societies · IEEE Photonics · RaeS · AIAA