How I got here.
A forward-chronological trajectory, from a mechanical-engineering undergrad and a single-propeller UAV, through a robotics master’s at TU Delft, to embodied-AI research, a string of hackathon wins, and building agritech robots in a castle in Austria.
- 2018-2022
EducationI started with mechanical engineering
NITK Surathkal
I came into robotics through mechanical engineering, powertrains, mechatronics, and the feeling that robots were the place where hardware, software, control, and people all met. I also started writing and publishing technical projects early, which later became a real habit.
- 2021
ResearchI published my first robotics paper independently
IEEE ICDTMRA
This one was not a polished lab pipeline. I wrote and published a TurtleBot3 localization and navigation paper with two juniors, without a professor driving the work. It was scrappy, but it mattered because it made research feel like something I could initiate myself.
View IEEE paper ↗ - 2018-2022
ProjectI kept building and writing in public
Hackster
My early Hackster projects were mechatronics-heavy and imperfect, but they crossed 100k+ cumulative views. That was one of the first signals that documenting technical work could create real surface area around what I was learning.
View Hackster ↗ - 2022-2024
EducationRobotics became HRI at TU Delft
MSc Robotics
At TU Delft, I finished the robotics master's early, took extra coursework, and kept getting pulled toward human-robot interaction. The question was already forming: if robots are complex, how do people actually understand and work with them?
- 2024-2025
ResearchMy thesis turned that into language interfaces
TU Delft / Frontiers
For my master's work, I built a hierarchical multi-LLM conversational interface for supermarket robots. It used intent classification, RAG, fine-tuning, speech interaction, and robot navigation. It was HRI, but it was also an early version of my current obsession: making robot behavior easier for humans to supervise.
Read the Frontiers paper ↗ - 2024
ProjectI shipped Unspool end-to-end
Solo product
Unspool was a voice journaling app, not robotics. Still, it mattered: I designed it, coded it, added auth/privacy/cloud sync, and got it through both app stores. It taught me the difference between having an idea and actually shipping the thing.
Visit Unspool ↗ - 2024-2026
WorkQafka gave me real deployment instincts
Robotics & AI Engineer
At Qafka, there was a robot, a hard physical problem, and an outcome to deliver. Without the usual academic padding, I had to make perception, control, state machines, embedded actuation, and safety logic work together. That changed how I think about robotics: reliability lives at the interfaces.
- 2025
Award / WinI used hackathons as fast robotics labs
Snap / FORGIS / EF / Junction
The hackathons were not the identity, but they were useful reps. I won at Snap, FORGIS, and EF, and built a range of things fast along the way: a Dobot pick-and-place demo, an AR Hyrox coach, and an elderly-care robot at Junction. They trained speed, taste, and the habit of getting a physical prototype to actually move.
- 2026
Work
Alpine Valley became the pivot
Founder in Residence
I tried building Latent Robotics around berry harvesting. We built a lot: SO-101 arms, teleoperation data, SmolVLA policies, HG-DAGGER, RL/MuJoCo experiments, and a fully automated fake-plant picking setup. But the sharper realization was not "I am an agri-tech founder." It was that embodied foundation models need interpretability and steerability before they can be trusted in the real world.
- 2026
ResearchThe interpretability blog made the bet explicit
Hugging Face
The strawberry-pick interpretability post was where the direction clicked: can we build observability layers around black-box robot policies, infer task stages, probe hidden representations, and eventually create handles for steering behavior?
Read the blog ↗ - 2026
ProjectI started contributing to LeLab
Hugging Face open source
LeLab became my first serious open-source robotics contribution path. The work is practical - bug fixes, features, setup, usability - but that is exactly why it matters. Better embodied-AI tooling makes experimentation easier to inspect and repeat.
View LeLab ↗ - Aug 2026
WorkNext chapter: Lely
Rapid Prototyping Engineer
At Lely, the role is broader than my independent research bet: rapid prototyping across electronics, software, AI/ML, mechanical systems, sensor fusion, and integrated prototypes. It is the professional next chapter; the interpretability and steerability thesis continues through my own research, writing, and open-source work.