systems online · Gurgaon, India
Technical Lead at Stashfin — I architect the platforms that put AI agents into production: routing engines, RAG pipelines and no-code tooling that turn weeks of engineering into minutes of configuration.
01 · proof of work
Every metric below is live or was measured in a shipped system. This is what my work does when real users hit it.
LLM tokens processed every day by the customer-support agent I architected — ~30,000 messages across 6,000+ daily users.
Requests/day through a centralized LLM routing engine — 15–20ms latency, 99% uptime, multi-provider fallbacks & circuit breakers.
D15 retention on an AI companion chatbot — LangGraph multi-agent planning + RAG memory.
Faster content production — promo generation cut from one week to 3 hours.
Agent delivery time after the no-code platform — prompt, tools, model & channel, all self-serve.
02 · systems
Platforms and infrastructure designed so that other people — engineers and non-engineers alike — can ship intelligence.
A self-serve platform for building and deploying production AI agents — configure prompt, tools, model and channels from a UI, ship in minutes. Runs multiple production agents today, including a support agent serving 6,000+ users daily.
agent delivery: days → minutes · 44M tokens/day in production
One gateway for every LLM call in the company — budget tracking, rate limiting, multi-provider fallbacks and circuit breakers so product teams never think about provider outages.
300K+ req/day · 15–20ms · 99% uptime
LangGraph multi-agent architecture for response planning and conflict resolution, RAG memory on Qdrant, and human-like dynamic response timing.
46% D15 retention
Turns any REST API into an agent-callable tool: configure the request, fire a live test call to capture the real response shape, annotate keys — the LLM tool schema writes itself. Zero hand-written integrations.
any API → agent tool, no code
End-to-end knowledge pipeline — upload, chunk, embed, store — plus a contact system that unifies a user's email, phone and Slack identities into one persistent memory, so agents never start cold across channels.
custom knowledge bases, zero engineering support
Parallel processing of 50–100 microseries for ad creative — accelerating A/B testing cycles and campaign deployment, with a no-code bot management console for product managers on top.
1 week → 3 hours production time
03 · timeline
FEB 2026 — PRESENT CURRENT
NodeJS · TypeScript · AWS · Postgres · GenAI · Docker · Microservices
JUL 2025 — FEB 2026
NodeJS · Postgres · Qdrant · MongoDB · LangGraph · GenAI
MAR 2022 — APR 2025
NodeJS · TypeScript · AWS · MongoDB · Kafka · Microservices
MAY 2021 — MAR 2022
NodeJS · AWS SQS · MongoDB · VueJS
OCT 2020 — MAY 2021
ReactJS · Angular · SQL · PHP
JAN 2019 — OCT 2020
Angular · C# · SQL
AUG 2015 — JUL 2019
8.7 CGPA
04 · stack
05 · about
I'm Sarthak — a Technical Lead based in Gurgaon, India. My career has one through-line: removing the engineering bottleneck between an idea and a running system.
At Bosch I learned discipline. At startups I learned speed. At Dresma I learned to build things that don't fall over. And in the AI era, I've found the work I love most: platforms that let non-engineers deploy production-grade AI agents — routing engines that survive provider outages, RAG pipelines anyone can feed, tool frameworks that turn any API into an agent capability.
I lead small teams, I stay close to the code, and I measure my work in production numbers — not promises.
06 · contact
Whether it's AI infrastructure, agent platforms, or a hard scaling problem — my inbox is open.