From curious experiments
to shipping AI products.
I'm Debanshu Devgoswami — an AI engineer who builds voice agents, LLM apps, and agent reliability tooling. I care about systems that work in the real world: clean APIs, honest evals, and products that actually ship.
The best way to learn AI is to build with it — ship early, measure everything, and refine until it works.
Location
Gujarat, India
Status
Open to opportunities
Focus
AI Agents · Voice AI · RAG
How it started
It started with a simple question in college: how does software actually understand us? That curiosity pulled me from coursework into AI — first machine learning, then LLMs, then agents that can actually do things.
My internships taught me how real teams ship. At AgentraTech AI, I worked on a voice-based loan agent that held real customer conversations and turned them into structured application data. At PrognosLabs, I built FastAPI automation that made campaign operations 60% faster — and learned that the unglamorous backend work is what keeps products alive.
Somewhere along the way I got obsessed with a harder question: how do you know when an AI agent is actually reliable? That became AgentGuard — my platform for evaluating agent workflows across reliability, tool-call correctness, and security. It's the project that taught me the most, because evals don't lie.
Now I build at the intersection of voice AI, agents, and reliability — the systems I wish existed when I started. I'm still early in the journey, but I'm doing it by building.
How I work
Build to learn
Tutorials fade. Working systems teach. Every concept I learn ends up as something running.
Measure, don't guess
Evals, tests, and benchmarks — if it isn't measured, it isn't working yet.
Ship the boring parts
Auth, rate limits, Docker, docs. The unglamorous work is what makes software real.
Stay curious
AI moves fast. I keep up by building, reading, and talking to people ahead of me.
Beyond the code
Outside of work, you'll find me reading about AI research, tinkering with side projects, or playing chess. I like learning in public and sharing what actually worked — and what didn't.
I'm drawn to builders who combine technical depth with clarity. That's the kind of engineer I want to become.