Vyomesh Bhatt / BITS Pilani, Goa / M.Sc. Mathematics + B.E. E&I
INTELLIGENCEUNDERCONSTRAINTS.
So far, that has meant a 10–12 KB model at Eureka Forbes, LiDAR vision for ICC broadcasts, a TinyML navigation smartglass, and RAG systems with actual token limits.
SCROLL TO INSPECT THE WORK ↓
Choose your lens
How much detail do you want?
01 / SIGNAL
I’ve worked across edge hardware, broadcast vision, robotics, and LLM infrastructure. The useful part is what each project had to get right.
Start with the numbers, then open a dossier if one matches the work you are hiring for.
The fast evidence brief
A few numbers worth
remembering.
These are from work at Eureka Forbes, Quidich, Sense Lab, and Project Kratos. They are here to give you a quick read before the details.
Selected work
Open one if you want
the important parts.
Problem, constraint, decision, result. That is the format. The work is easier to judge when the trade-offs are on the page.
Find the closest match
What kind of problem
are you hiring for?
Pick a few conditions. I’ll point you to the closest examples from the work above.
SELECT ONE OR MORE CONDITIONS
Choose a condition to get started.
The matches below lead back to a company, lab, or system in the record.
Where the tools show up
Every tool has a
specific place.
Click a technology to see the project it belongs to. No skills matrix, no floating buzzwords.
CLICK A TECHNOLOGY
Project first. Tool second.
The work stays understandable even when the tool changes. The project is the useful context.
Pick a tool to see where it was used and what it was responsible for.THE TECHNOLOGY MAP
A recurring shape
The projects are different.
The handoff is similar.
Signal comes in. It gets cleaned up and structured. A model makes a decision. Someone—or another system—has to trust the result.
Purifier telemetry, camera + ToF, LiDAR, IMU, documents.
DSP, fusion, OCR masking, topology extraction, retrieval.
TinyML, CV, PPO, RAG, function calling.
Precision, false positives, citation-backed output, fallback logic.
MCU, broadcast, rover, workflow, person in the loop.
Most failures happen at the handoff. That is what I look for.
Easy to check
Three public signals
worth opening.
Navigator
at EWSN 2025
An ESP32-S3 TinyML smartglass combining camera and time-of-flight sensing for real-time indoor navigation assistance.
95% TPR · zero false positives · sub-300 mssktime
contribution
Implemented the HampelFilter detector, optimized RelativeLoss evaluation throughput, and improved QA coverage for MovingWindow and ClaSPTransformer.
EUROPEAN SUMMER OF CODEProject Kratos
Mars Rover
Led a 50+ member perception and navigation team across closed-loop autonomy, mapping, localization, and human override.
ROS · YOLOv5 · CUDA · RTK-GPSIf this overlaps with your work
Edge AI, perception, robotics, or LLM systems? Let’s compare notes.
I’m looking for a team where I can take an ML problem from signal quality to model behavior, system interfaces, and validation.