Backend engineer. I own production systems end to end: the migration, the billing run, the 2 AM root cause, and the LLM pipeline clinicians rely on the next morning.
Current: Senior Software Engineer at Stack AI Solutions, embedded in a US telehealth platform serving 74k+ patients, 400+ providers, and 100+ facilities (HIPAA).
Most of my best work lives in private client repos, so here is the short version with numbers:
- Re-platformed a live healthcare system from legacy PHP/MySQL to FastAPI + PostgreSQL and ran the production migration myself: 60+ idempotent scripts, incremental cutoff sync, zero data loss. The platform now manages 110k+ visits.
- Built the end-of-month RTM billing engine: centralized CPT charge calculation, bulk e-signing, and an Excel dry-run a human verifies before every run. Billing code must show its work before it moves money.
- Cut a core endpoint from 18.5s to 5.6s with set-based batching and pg_trgm indexes. Root-caused recurring full-server OOM crashes (workers at 13.6 GB on unbounded payloads) and eliminated the outage class.
- Ship production LLM systems, not demos: 3,000+ automated SOAP notes and 2,300+ patient summaries in daily clinical use; an exam-grading microservice built near-solo (GPT-4o vision, LiteLLM, Langfuse) with a grader-plus-auditor agent design and a must-quote-the-source anti-hallucination rule.
- Automate the annoying by default: an internal ticketing tool built solo in ~3 weeks (1,000+ users, $50K+/year saved), a Claude pre-push deploy gate, a daily autonomous bug-triage pipeline.
Stack: Python, FastAPI, PostgreSQL, TypeScript, React, AWS, Docker · LiteLLM, Claude API, GPT-4o, Langfuse, RAG, agent design, LLM evals, structured outputs, prompt engineering
Earlier: Nagarro (cut ~$1M/year in cloud costs migrating a central DB to Azure Cosmos DB) · IIIT Bhopal, CSE '23