I build data platforms & AI/ML pipelines that run faster, scale further, and cost less.
7+ years across data platform, backend, and ML systems. I rearchitect heavy AI/ML pipelines, move workloads to Kubernetes at scale, and own ambiguous, high-stakes platform problems end to end — from Spark-on-GPU jobs to crawlers processing tens of millions of domains.
Six problems I owned end to end. Each one started ambiguous and shipped with a measurable result.
Overhauled Demandbase's AI-driven named-entity-recognition pipeline (spaCy on Spark, GPU-accelerated). Runtime collapsed from 56 hours to 3 while cutting compute spend 75% — turning an overnight batch into a same-day job.
Moved the Render Crawler off AWS Lambda onto Kubernetes: 8× domain-processing throughput, 33M domains in 5 days, and monthly cost down from $10K to $2K (80% reduction).
Recovered roughly $96K/year in S3 (~$8K/month) through proactive data upkeep, and geotagged 4.5M locations with Pelias instead of Google APIs — a 95% cost cut ($10K → $500) with no loss in quality.
Took ownership of a fragile data product and cut upgrade time 90% — from 4 days to 30 minutes — alongside catalog migrations, EMR upgrades, and security hardening in a company-wide shift-left program.
Boosted search processing 1,000% and price-cache performance 200%, cut API latency 80%, shrank Redis keys from billions to thousands, and built APIs handling 50M daily updates.
At Odessa, automated regression testing and releases to bring delivery time from one week to five hours.
From OCR support and release automation to owning AI/ML data platforms — a steady climb toward harder, higher-leverage systems.
Award-winning internal AI work, open-source tooling for the messy realities of AI-generated code — and a lab experiment you can run right on this page.
One 1-layer Transformer trained on single-digit +, −, ×, ÷, rebuilt as three 512-resistor crossbars (ngspice-validated), then composed by a digital controller into a full multi-digit calculator. Type any numbers and press =; every single-digit step runs through the real circuit, the way a CPU builds big math from small.
A static-analysis CLI built for the specific failure modes of AI-generated Python. 40+ rules targeting AI-prone bugs, framework-aware checks for FastAPI, Django, and Flask, and SARIF output that drops straight into CI/CD.
The project that won Demandbase's Velocity Hacker prize at the September 2025 AI: Show & Tell — recognized as the standout build among company-wide entries.
An AI-powered outreach assistant that uses ML to surface high-potential leads and automate personalized outreach — Most Impressive Technical Implementation at the Demandbase AI Hackathon.
Targeting senior data platform, backend, and ML platform roles — remote-first from India.