A kid who screws and unscrews things to see how they worked. The screwdriver just got a GPU.
Hi there! I’m Durgesh, an AI/ML engineer based in Bengaluru with a freshly minted M.Tech in Machine Intelligence (PES University, CGPA 9.41, Prof. CNR Rao Scholar, which I’ll take as proof that the late-night debugging sessions were, on balance, worth it). Curiosity has always been my first instinct, and taking things apart my first method. I still can’t leave a black box closed, and I’m happiest when it’s messy, real and slightly misbehaving.
My first big box was at ISRO’s ISTRAC, where I spent six months engineering an NLP pipeline that gave scientists 10 times faster access to satellite telemetry data. That felt appropriately high-stakes for someone who still double-checks whether a for-loop is off by one.
Most recently, at the Indian Navy’s AI Incubation Centre (INICAI), I built a real-time multi-camera person re-identification system on NVIDIA DeepStream. That’s a fancy way of saying I taught a network of cameras to stop losing track of people every time they walked behind a pillar.
The pattern across all of it is simple: find the part that isn’t working, open it up, understand it, and make it hold up under real conditions. It’s why I’m drawn to the gap between AI research and what actually survives production: reliability and real impact first, elegance second (though I’ll take elegance if it’s free).
Around the edges, I lead hackathon teams, mentor others, contribute to Open Source (Simple Icons), solve an unreasonable number of LeetCode problems at hours polite people are asleep, and build the odd side project, like the one you’re looking at right now.
Off-screen, I stay in motion: years of Taekwondo before I hung up my dobok, and I’ll never say no to cricket or volleyball (the patience for a stubborn bug and for a stubborn opponent come from the same place). The rest of my time splits between trails and trees, anything that isn’t a screen, and the cosmos, which is really just nature that happens to be very far away. This site’s background quietly swaps between the two, which is less a design choice and more a confession.
The working record: internships, open-source commits, degrees, certifications and the communities I learn alongside. Everything here is something I can point to, with numbers wherever there are numbers, and it’s going to keep growing.
Built a real-time multi-camera person tracking & re-identification system that keeps a consistent ID on someone across CCTV feeds — the same core problem FoundYou solves, this time for the Navy.
Wired Whisper and a fine-tuned RoBERTa into voice/text command systems for mission control, then built a Whoosh + spaCy semantic search engine that got scientists to satellite telemetry data 10 times faster.
Contributed several brand icons and updates to a widely used open-source icon library, keeping every change JSON-schema compliant through standard GitHub PRs.
Specialty: Software Engineering, Machine Learning and Artificial Intelligence.
Relevant coursework: Deep Learning Theory and Practices, Stochastic Models and Machine Learning, Speech and Natural Language Processing, Distributed Computing, Scalable Computing.
Specialty: Software Engineering and Data Science.
Relevant coursework: Machine Learning using Python, Principles of Artificial Intelligence, Introduction to Deep Learning, Big Data Analytics, Information Retrieval and Organization, Calculus, Linear Algebra and Optimization Techniques.
Merit Certificate + 40% tuition-fee refund, awarded to the top 20% of students by SGPA. Named for India's own Bharat Ratna chemist — not a bad namesake to share a merit list with.
A close-up look at how real missions plan and run spacecraft operations — attended during my ISRO ISTRAC internship, which made the mission-ops talks land less like theory and more like the building next door.
A developer-focused deep dive into Guidewire’s insurance-platform engineering — a good reminder that "boring" industries like insurance run on genuinely hard distributed-systems problems.
OpenAI’s first DevDay Exchange in India — a front-row look at agentic AI, from GPT-6.1 Sol and Codex to the Agents API. Left convinced software is less "write every line" and more "delegate it well."
Hackathons are where I find out whether a plan survives a deadline: three times leading the team, and once mentoring one from the sidelines. The result comes first, the project underneath, and the scoreboard is the only thing in this section that doesn’t negotiate.
People whose work makes me want to build something that actually holds up — across security, product, AI research, and physics. Not a complete list, and it’s going to keep growing.
Built RSS at 14, then spent his life arguing that knowledge shouldn’t need a paywall. A reminder that code can be an act of principle, not just a job.
Proof that the most interesting vulnerabilities are rarely in the code — they’re in the assumptions everyone else stopped questioning.
Obsessive about the gap between "it works" and "it feels right" — the same gap I chase between a notebook model and something people can actually rely on.
Asked whether machines could think before "machine learning" was a phrase anyone used — the original ask-the-obvious-uncomfortable-question move.
Kept betting on neural networks through decades where that bet looked unfashionable. Conviction, backed by math, eventually wins.
An apple, a law, and centuries of engineers still building on top of it — also the entire reason the Newton-Gravity project exists.
Thought in systems decades ahead of the hardware needed to build them. Vision without the ability to execute is still worth something.
Did field-defining work with a fraction of the resources and none of the institutional goodwill her male peers had. Rigor as the great equalizer.
Discipline off the keyboard: years of Taekwondo, and a standing yes to cricket or volleyball. The patience for a stubborn opponent turns out to be the same patience a stubborn bug asks for, just with more cardio.
Trained as a Taekwondo player — the kind of discipline where "practice the fundamentals until they’re boring" turns out to be the actual secret, same as with code.
A lifelong cricket fan and player, for the same reason debugging is oddly satisfying: most of the game is patient observation, then one sharp decisive moment.
Plays volleyball for the forced teamwork — no single player wins a rally alone, which is a healthy thing to remember after too many solo side projects.
Pick your protocol — email, LinkedIn, GitHub, X, or YouTube. All return 200 OK.
durgeshgowdac.