Research

Four themes, one approach: build the sensor or dataset, then the model, then put it in front of the people who need it — and release the tools so others can build on them. All publications.

Clean Air

Most Indian cities have few air monitors. We build low-cost sensing, models that fill the gaps between monitors, and LLM tools that let non-experts ask questions of the data.

AirDelhi NeurIPS ’23 Scalable Gaussian processes AAAI ’22 Sensor placement AAAI ’26 VayuBuddy & VayuBench CODS ’25 Samachar COMPASS ’22

Health Sensing

Measuring lung function or calories burned usually needs a clinic. We turn thermal cameras, face masks and wearables into measurement tools, and test how well AI models read them.

JoulesEye IMWUT ’24 SpiroMask TOCH ’23 ThermEval KDD ’26 Signal descriptors DSAA ’26 ApneaEye patent filed ’25

Energy

An electricity bill says how much a home used, not what used it. Since 2014 we have built the open toolkit, benchmarks and models the field uses to answer that.

NILMTK Test-of-Time Award ’25 NILMBench2026 Best Paper Runner-Up ’26 “I do not know” BuildSys ’22 Gemello KDD ’16

Earth Observation

Polluters such as brick kilns are spread over areas too large to survey. We map them from satellite imagery and connect detections to compliance rules.

SentinelKilnDB NeurIPS ’25 Space to Policy JCSS ’24 FlowSat BMVC ’26 PlantDoc CODS ’20

Methods & explainers

The machine learning underneath — Gaussian processes, active learning, uncertainty — and interactive articles that teach it.

Exploring Bayesian Optimization Distill Active Learning: A Visual Tour VISxAI Exploring Hidden Markov Models VISxAI

Datasets & tools you can use