AI/ML Engineer building production-grade GenAI, RAG, and agentic systems — from LangGraph multi-agent pipelines to full-stack web apps, with a growing obsession with system design.
I'm a final-year B.E. student in AI & ML at CMR Institute of Technology, Bengaluru, currently working as an AI Developer at Involynk, a Bengaluru-based startup, where I build agentic and LLM-powered systems for real product use cases.
My work spans the full stack of applied AI: retrieval-augmented generation, multi-agent orchestration with LangGraph, and shipping the FastAPI/Django backends and React frontends that make those systems usable. I favor small, well-commented, explainable systems over black boxes — the kind of project I can walk through end-to-end in an interview or a demo.
Before this, I interned as a Django Developer at Elevance Skills. Outside of shipped work, I spend time on physique training and iterative self-improvement — the same disciplined, step-by-step approach I bring to building software.
There's something I've never quite gotten over: someone I will never meet opens an app I built, at some hour I'll never know about, and it just works for them. They don't know my name. I still, somehow, love being called "the programmer."
Building AI agents and AI automation for production use cases, working with RAG, Open Knowledge Format, and developing and scaling full-stack applications — from system design to deployment — ensuring they can scale using Kubernetes, then wrapping it all in Docker so it can pretend to be someone else's problem in production.
Developed and maintained backend features on a Django-based platform.
Coursework spanning RNNs, GRUs, LSTMs, probability, calculus, and competitive programming.
A resume is one page. My brain, unfortunately, is not. Here's the rest of the stack I actually work with day to day — presented here because Word wouldn't let me shrink the font any further.
Keeping prompts, latency, and hallucination rates from ruining someone's Monday standup.
Docker, CI/CD, and just enough YAML to question every life decision that led here.
Whiteboard architectures that mostly survive first contact with production traffic.
Turns out "it felt right in the demo" is not a valid evaluation methodology. Who knew.
Because interoperable, structured knowledge shouldn't need a PhD to parse.
LoRA, QLoRA, and the eternal question of whether the GPU bill was worth the 2% eval bump.
Getting multiple agents to cooperate without starting an argument in the logs.
Multi-agent systems that do what I intended roughly 80% of the time — which is apparently industry standard.
No-code automation, for the days writing actual code feels like too much code. But works well 😁
Wiring tools into Claude so it can pretend to check my calendar and act busy on my behalf.
If I'm honest, more Redis than actual database at this point — I just can't wait for a query to hit disk like some kind of animal.
Because letting the model decide everything unsupervised felt like a great way to end up in an incident report.
Backprop, gradients, linear algebra — proof the degree wasn't wasted, dusted off under duress in interviews, then immediately forgotten again by Friday.
Corporate-speak for "47 browser tabs open, three courses half-finished, and a deep personal relationship with documentation I'll read later."
Kafka, RabbitMQ, Celery — decoupling services one queue at a time, one dead-letter topic at a time.
Advanced proficiency. Unpaid. Ongoing field research into AI discourse and hot takes. Not on the resume, but arguably my most-used skill this week.