What I built, where, and why it mattered.

Remote

Founding engineer on Lumen, a healthcare claims intelligence SaaS platform — employee/contractor number one, before any other technical hire. Built the entire system from an empty repository to a live product with real paying healthcare organizations using it daily, then hired and onboarded the team that maintains it today.

• Joined as the founding engineer with no existing codebase, architecture, or tooling — defined the technical direction, stack, repository structure, and conventions the whole team now follows.
• Owned the full stack end-to-end: Next.js/React/TypeScript frontend, Django/DRF backend, GCP infrastructure, data pipeline, billing, and RBAC.
• Designed a dual-database architecture with a custom Django database router — application data and scraped claims data live in separate, independently-managed databases, fully decoupling the data layer from the application layer.
• Built a complete Stripe billing lifecycle from scratch: trial provisioning, plan upgrades/downgrades, webhook idempotency, and automatic suspension/restoration of users and connectors tied to plan limits.
• Identified and enforced a critical multi-tenancy invariant (a key record identifier was not globally unique across accounts) — designed and documented the account-scoping rules across every query path to prevent cross-account data leaks.
• Built an automated data collection pipeline with post-collection QA gating (row-count, null-rate, and duplicate checks) before data reaches production, with Slack alerts on failure.
• Hired, onboarded, and mentored the engineers who followed — frontend, QA, DevOps, and backend/integration — including project memory systems and AI-assisted developer tooling that got new engineers contributing correctly from day one.

Bengaluru, India

Architecting and scaling the core LLM processing infrastructure behind AI Hyper Cube — the platform that runs scoped LLM prompts over AI-generated search content to extract intent, brand entities, sentiment, and citations at production scale.

• Architected and scaled the core LLM processing infrastructure for AI Hyper Cube using Python, ClickHouse, and MySQL, processing 5B+ tokens per month.
• Engineered a GPU-based LLM inference platform using vLLM, Vast.ai, and open-source models for large-scale entity extraction, cutting monthly AI processing costs 62% ($40K → $15K).
• Built pre- and post-processing pipelines for AI Hyper Cube, processing 100M+ keyword SERPs using Python and BigQuery for prompt generation, and Spark for post-processing with brand-level hash partitioning (MD5 modulo).
• Built a scalable Trino–Iceberg–ClickHouse data pipeline with a custom memory-aware job scheduler, replacing the legacy BigQuery UDF process — $18K–$20K in monthly savings while processing 1B+ events per month.
• Led development of DCX Collector V2 (Python, FastAPI, RabbitMQ, Redis, Docker, Argo Workflows, BigQuery) — a 4x faster processing pipeline handling 100M+ keywords per month across multiple vendor integrations.
• Implemented an automated AI Overview stitching system (Python, JavaScript, BigQuery UDFs, Argo) that mitigated the Google num=100 deprecation and cut keyword-collection costs ~60% across 25+ locales.

Bengaluru, India

Owned backend APIs and frontend experiences end-to-end for a data-pipeline SaaS product — from revenue-driving integrations to Core Web Vitals performance work.

• Engineered the Snowflake Partner Connect feature via a secure POST API with a JWT fingerprint for automated destination creation, driving an additional 10–15 MQLs per month.
• Optimized the Signup API, cutting response time from 9s to 2–2.5s (~75% decrease) and increasing signup conversion 4–5%.
• Led end-to-end development of a secure referral portal (JavaScript Promises syncing job submissions across the database, webhooks, and the Lever API), increasing referral engagement 40%.
• Migrated the Django website frontend to a Next.js app with TypeScript and Tailwind CSS, and Dockerized it — a 15% improvement in application load times.
• Implemented A/B testing on high-conversion pages (signup, demo) using Jinja templates, HTML, SCSS, and JavaScript, resulting in a 20% increase in demo requests.
• Enhanced Core Web Vitals by deploying Varnish in-memory caching (good LCP URLs 55.73% → 65.23%) and migrating to CloudFront CDN, further raising good LCP URLs to 83.85%.

New Delhi / Bengaluru, India

Early-career work spanning ETL pipelines, Python automation, and evaluating machine-learning APIs for production fit.

• Created an ETL pipeline using Google Apps Script that fetched real-time data from a stock API and loaded it into Firebase and Google BigQuery — a 20% reduction in project costs.
• Evaluated 10+ machine-learning APIs, including the Google Video Intelligence API, and improved backend architecture with Python and MySQL, achieving ~90% accuracy in video analysis.
• Reduced project costs 15% by scripting Python-based data seeding for Google-registered shops via the Google Places API into a Firebase database.
• Developed Python scripts automating Confluence-to-Zendesk knowledge-base uploads, improving workflow efficiency for the customer experience team.