Shipping production systems across fast-moving startups — including a unicorn. Full stack, cloud infra, and now production AI agents that enterprise teams depend on daily.
Not a résumé. Just the honest version.
I'm Rethesh — a Senior Software Engineer from Hyderabad who's spent his career learning what it actually means to build software in environments where things move fast and mistakes are expensive.
It started during college — freelancing for clients while still studying, shipping websites, event platforms, and web experiences for brands I'd have been too nervous to approach a year earlier. That hustle is still in me. The urgency of figuring things out quickly never left.
Then came the corporate world — not the slow, comfortable kind, but startups that were scaling hard and needed engineers who could own things end-to-end. A D2C unicorn. An AI-backed edtech company. Each one taught me something the previous one couldn't: how engineering decisions made under speed affect teams for years.
Right now, I'm deep in agentic AI systems — building things that feel genuinely new. Conversational interfaces that reason over complex infrastructure data, query databases in natural language, and reflect on their own outputs before answering. Shipped for 50,000+ enterprise users across the career. This is the most technically interesting work I've done, and I'm only getting started.
Full stack in the truest sense — from schema design to deployed cloud infra, from CI pipeline to AI agent graph.
High-level. The full story is in the résumé — or in a conversation.
Building enterprise agentic AI on a cloud infrastructure platform — GPU cost optimisation, real-time monitoring pipelines, and a conversational AI assistant for infrastructure analytics. Leading a small team while shipping complex systems end-to-end.
Led engineering on a corporate LMS serving 50,000+ enterprise users. End-to-end ownership of features, AWS infrastructure, and platform security within a unicorn-backed AI company. Moved fast without breaking the things enterprise clients depend on.
Contributed to 8 internal platforms at a D2C unicorn — MDM, B2B, Warehouse, Finance, Content, Reports, and more. Built the Angular Material component library adopted org-wide. This is where I learned what engineering at unicorn scale actually feels like.
While still in college, shipped web experiences for 8+ clients across very different domains — event platforms, music & media websites, e-commerce storefronts, promotional campaign sites, and brand landing pages. Each project was a different problem, a different constraint, a different lesson. This period built the instinct to figure things out fast and ship something real.
The problem space that keeps pulling me back after hours.
What if anyone could query a database just by asking a question — and get a reliable, grounded answer back?
My work on production AI systems pushed me deep into this problem. When an enterprise user asks "what were our top GPU cost drivers last month?" — that natural language query has to be translated into MongoDB aggregations, executed accurately, and explained back in plain English. Getting that right in production, at scale, with no hallucinations, is genuinely hard.
This has led me into active research around NL-to-SQL and NL-to-NoSQL query generation — using LLMs with schema awareness, few-shot prompting, and reflection loops that catch errors before they reach users.
I'm also exploring LLM-powered analytics pipelines — cron-driven data collection feeding vector stores, with conversational interfaces layered on top. The infrastructure exists. Making it trustworthy at enterprise scale is the hard, interesting part.
Ran an empirical study evaluating three prompting strategies for NL-to-SQL translation on the Chinook database using gemma3:4b via Ollama — fully local, no API. The result: schema injection alone jumped execution accuracy from 0% → 92%, and an error-driven retry loop closed the remaining gap to 100% across 50 structured queries. If you want some info about this, it's all here.
A couple of things I built outside of work — the kind that sharpen thinking differently than production pressure does.
I write when something is worth sharing — usually something learned the hard way.
Not everything I've done has been code. Alongside the engineering work, I've volunteered, run campaigns, and collaborated with some interesting organisations — from national media brands to cycling communities. These gigs kept me connected to the world outside a terminal window, and made me a better product thinker because of it.
"The curiosity I bring to a long bike ride is the same one I bring to a hard engineering problem."
When I'm not in front of the screen, chances are I'm out on my Triumph Scrambler — full gear on, friends alongside, no particular destination. Weekend rides with the crew are sacred. There's something about being on a bike that forces you into the present moment in a way a screen never can.
Recently picked up pickleball and it's been properly addictive — fast, social, and surprisingly technical once you get past the basics. Pair that with exploring Hyderabad's food scene and hunting down good coffee, and most weekends are accounted for.
Cricket is still in the blood. Grew up with it, still playing. Team sport maps cleanly to engineering — everyone has a role, communication is everything, and outcomes depend on collective execution.
I travel when I can, and I think it makes me a better engineer. Good products are built by people who've lived outside their desks. The AI space right now is the most exciting moment in software I've lived through — and I intend to stay at the front of it.