Data that holds up.
Systems that ship.
Based inCASE STUDIES
1.99M rows of FMCG supply-chain data turned into forecasts a distributor can actually order against.
Medallion pipeline turning 1.99M raw rows into orderable forecasts
Radar that finds a heartbeat through a wall — built so a fire crew can carry it into a burning building.
Through-wall vital-sign detection for disaster response teams
A digital twin of a railway section that decides which train moves next, and an Android app to watch it work.
Digital twin deciding which train moves next across a section
5M+ Aadhaar records read for the story they tell — district stress, demand spikes and migration that shouldn't be there.
Predictive governance over 5M+ Aadhaar ecosystem records
View system details
An RL environment where two trains structurally cannot occupy the same block — safety encoded, not rewarded.
Safety encoded as an invariant, not learned from reward
Spatial attention and Grad-CAM heat maps showing exactly where a model looks in an aerial frame.
Spatial attention and Grad-CAM interpretability on aerial imagery
Challenge the problemdeliver with precision
Most of what I build starts as someone else's unglamorous problem — a zero that means two different things, a signal buried in clutter, a rule that has to hold every time.
KNOW ABOUT ME
I'm Omkar — a Computer Engineering undergrad in Pune who spends most of his time on the unglamorous half of machine learning: the datasets, the pipelines, and the automation that keeps them honest.
Three national hackathon placings in eighteen months, a year of production dataset work for autonomous-driving and retail AI clients, and a standing interest in Linux internals, reverse engineering and anything that can be scripted.
1.99M
records engineered
10.4%
forecast MAPE
320
outlets modelled
5M+
records analysed
FIELD NOTES
Ask about OmkarAIAI assistant · from his résumé
An AI assistant, answering from Omkar's résumé.
Projects, roles, stack, availability
