I ship LLM-integrated products and the systems that build them, fast.

A multi-tenant WhatsApp booking SaaS for salons and spas — businesses take bookings, enquiries, and payments entirely over WhatsApp, no app install. I built it from scratch: tenant isolation on a shared TiDB (MySQL) database, a per-tenant Twilio number, a pure-function FSM bot (8 states) with a Gemini info-bot, Razorpay payment links with HMAC-verified webhooks, and a Next.js dashboard. Launched commercially, onboarding first paying tenants.

outbuiltit.comopen ↗

stack Next.js 16 · TiDB Cloud · Twilio · Razorpay + HMAC · Gemini · Prisma · 79 Vitest tests

A travel-planning platform that tells you when to visit attractions based on crowd-density and wait-time signals. Sole developer — designed the data model, built the frontend and API layer, deployed on GCP. Migrating the stack from React + PHP to Next.js (SSR + API routes) grew daily organic users from 60 to 400+, with zero paid acquisition.

crowd-tracker.comopen ↗

stack Next.js · MySQL · GCP · SSR

Software Engineer II

Firestorm Internet

• 01Built and run an end-to-end AI automation pipeline that scaled WordPress site production from 5 to 32 sites in 4 weeks — generating ₹1L+ in affiliate commissions in the first month. Handles AI content generation, image sourcing, and WordPress deployment, with a human gate only on image review.
• 02Designed a multi-agent content architecture after hitting single-agent failures — separated content creation (Claude) from HTML/template generation (Python + Claude); quality jumped once each agent had one responsibility.
• 03Built an async job queue (Celery + Redis) — every long task runs as a durable worker with retry policies, exponential backoff, and dead-letter queues. Treats LLM output as untrusted: structured-output validation, retries on malformed responses, human gates where AI is least reliable.
• 04Built an internal contact-management tool (React + FastAPI) aggregating enquiries from 40+ sites with a Gmail-integrated triage interface. Own containerization and CI/CD — Docker, Google Cloud Build, deployed on GCP.

Associate RPA Developer

UiPath

• 01Helped train and deploy an ML email-classification system (UiPath Communications Mining) routing 10K+ emails/month across 200+ client-configured sources — 85–90% accuracy, built with an AI scientist.
• 02Integrated classification outputs with SAP and ServiceNow for downstream routing and validation; redesigned exception handling and routing logic to improve workflow throughput by 70%.
• 03Implemented UiPath Computer Vision for remote-desktop automation — extraction and form-filling across legacy enterprise interfaces.