Research

Advancing AI for physical systems.

At Purnabodha Labs, every product begins as a research problem. We investigate machine learning methods that respect the temporal, physical, and spatial structure of real-world systems, transforming scientific ideas into deployable intelligence.

Active Research Programs

Five active scientific directions starting from the physical and mathematical structure of each domain.

Learning Physical Dynamics

Research Question

"How can neural networks model long-term temporal behaviour in real-world physical systems?"

Overview

Physical systems exhibit long-range dependencies, non-stationary behaviour, and interactions across multiple timescales. Our research explores architectures capable of learning these dynamics while remaining robust under changing operating conditions. We investigate representation learning techniques that transfer across industrial, environmental, and energy domains.

Current Directions

Long-context sequence modeling
Temporal representation learning
Continuous-time modeling
Forecasting under distribution shift
State-space sequence models for temporal modeling

Potential Applications

Long-horizon forecasting for energy and climate systems
Real-time equipment health monitoring
Degradation trajectory prediction for industrial assets

Multimodal Physical Intelligence

Research Question

"How can heterogeneous sensors be fused into a coherent representation of the physical world?"

Overview

Physical environments are monitored by diverse sensing modalities that operate at different frequencies, resolutions, and levels of reliability. Naive fusion methods fail to capture the complex spatial-temporal correlations between these streams. We research multimodal fusion architectures that align heterogeneous data and maintain robustness even under sensor degradation or failure.

Current Directions

Sensor fusion
Cross-modal attention
Missing sensor robustness
Representation learning

Potential Applications

Industrial sensor fusion for predictive maintenance
Environmental monitoring from heterogeneous sensor networks
Multi-view defect detection in manufacturing

Scientific Machine Learning

Research Question

"How can physical laws improve learning and generalization?"

Overview

Purely data-driven models frequently make physically impossible predictions when operating outside their training distribution. Fusing neural networks with conservation laws, thermodynamics, and electrochemical principles forces models to respect physical boundaries. Our research develops hybrid architectures that integrate domain knowledge directly into machine learning frameworks to improve data efficiency and safety.

Current Directions

Physics-informed learning
Hybrid models
Neural differential equations
Domain constraints

Potential Applications

Battery degradation modeling with electrochemical constraints
Turbine and rotor dynamics simulation
Material fatigue prediction under cyclical loading

Reliable AI for Physical Systems

Research Question

"How can AI know when it is uncertain?"

Overview

In safety-critical physical systems, overconfident predictions under unfamiliar situations can lead to catastrophic failures. Standard machine learning models do not inherently represent their own limitations under domain shift. We study probabilistic methods and calibration techniques that quantify predictive uncertainty, allowing AI to safely trigger human intervention or conservative control defaults.

Current Directions

Probabilistic forecasting
Calibration
Distribution shift
Risk-aware decision making

Potential Applications

Safety-critical forecasting in aerospace and energy
Uncertainty-aware autonomous inspection
Risk-calibrated predictive maintenance scheduling

Spatial Intelligence

Research Question

"How can AI systems understand the physical world through visual, multi-spectral, and spatial sensing?"

Overview

Interpreting physical environments from pixels requires going beyond generic image recognition to model geometry, multi-spectral bands, and temporal changes. Our research explores architectures capable of learning robust spatial representations from satellite observations, aerial imagery, and industrial visual data under varying environmental conditions to track environmental changes and assist in disaster response.

Current Directions

Spatiotemporal scene representation
Multi-spectral representation learning
Geometric priors for spatial reasoning
Foundation models for environmental monitoring

Potential Applications

Satellite-based environmental monitoring and change detection
Disaster response planning from aerial imagery
Infrastructure inspection from drone footage

Research by Application

Each application area presents unique scientific challenges. The following research directions are specific to each domain and evolve as our understanding deepens.

Grid Energy Intelligence

Current research directions for energy demand forecasting and grid optimization.

Renewable Integration Forecasting

Predicting demand under high renewable penetration with intermittent generation patterns from solar and wind sources.

Extreme Event Forecasting

Rare but critical grid events including heatwaves, cold snaps, and unplanned outages that challenge conventional load forecasting.

Cross-Region Transfer Learning

Adapting models trained on data-rich states to data-sparse regions with different consumption patterns and infrastructure.

Real-Time Adaptive Recalibration

Continuously updating prediction intervals as new smart meter data streams arrive, maintaining calibration under changing grid conditions.

Pollution Intelligence

Current research directions for air quality forecasting and environmental monitoring.

Satellite-Assisted Pollution Forecasting

Fusing columnar NO₂ and aerosol optical depth from Sentinel-5P and MODIS with ground-level monitoring networks for wider spatial coverage.

Large-Scale Representation Learning for Environmental Monitoring

Pretrained on global air quality data, adaptable to local monitoring networks with sparse station coverage and varied sensor configurations.

Spatiotemporal Representation Learning

Learning latent transport dynamics from sparse monitoring stations to reconstruct continuous pollution fields across urban areas.

Extreme Pollution Event Forecasting

Predicting pollution spikes from wildfires, dust storms, and temperature inversions with lead times sufficient for public health advisories.

Battery Intelligence

Current research directions for battery degradation modeling and RUL prediction.

Physics-Informed Degradation Models

Embedding electrochemical constraints into learned representations for reliable extrapolation beyond training regimes and cycle configurations.

Fast Adaptation to New Chemistries

Few-shot transfer learning across battery formulations, form factors, and manufacturing batches with minimal retraining.

Cell-to-Pack Representation Learning

Scaling cell-level predictions to module and pack-level remaining useful life accounting for thermal and electrical coupling effects.

RUL Under Unseen Operating Conditions

Generalizing across temperature, C-rate, and depth-of-discharge profiles not encountered during training.

Predictive Maintenance

Current research directions for industrial telemetry and equipment health monitoring.

Large-Scale Representation Learning for Industrial Telemetry

Pretrained on diverse sensor modalities, adaptable to new equipment types and operating conditions with minimal fine-tuning data.

Multimodal Maintenance Intelligence

Integrating vibration, thermal, acoustic, and operational logs for holistic equipment health assessment across disparate sensor networks.

Zero-Shot Fault Diagnosis

Identifying novel failure modes without prior examples, enabling early detection of unforeseen degradation patterns in field-deployed fleets.

Continual Learning for Evolving Fleets

Adapting to fleet-wide degradation trends and equipment refresh cycles without catastrophic forgetting of previously learned failure modes.

Open Science

Publications & Open Science

We believe research creates the greatest impact when it is reproducible and shared. Initial publications and open-source implementations will appear here as our research matures.

Get In Touch

Collaborate with Us

We welcome collaborations with researchers, universities, startups, and industry partners interested in advancing AI for physical systems.

Collaborate