AI Research
Edge AI for energy infrastructure. Intelligence that runs where the infrastructure is - on meters, on transformers, at the grid edge. From NILM to VLA model quantization.
The prevailing AI narrative is dominated by large-scale cloud models. But critical infrastructure cannot depend on round-trip latency to a data center. Transformers, smart meters, and industrial controllers need intelligence that is local, lightweight, and reliable - models that can run on milliwatts, not megawatts.
SARAL's AI research operates at this intersection: building inference pipelines that are small enough for edge hardware, accurate enough for operational decisions, and robust enough for deployment in environments where failure has real consequences. Our work spans non-intrusive load monitoring, energy theft detection, knowledge preservation through conversational AI, and adaptive quantization for vision-language-action models.
Featured Publications
Edge AI for Appliances & IoT
Deploying intelligence at the point of consumption. This research explores lightweight inference models for appliance identification, anomaly detection, and energy optimization running directly on IoT hardware.
NILM (Non-Intrusive Load Monitoring) Research
Disaggregating total household or facility power consumption into individual appliance signatures without sub-metering. SARAL's NILM research pushes accuracy boundaries on Indian load profiles.
All Publications
VLA Model Quantization (Adaptive Precision Inference)
Making vision-language-action models run on constrained edge hardware through adaptive precision quantization - reducing model size without sacrificing task-critical accuracy.
KTBot - Tribal Knowledge Preservation
Capturing and structuring the tacit knowledge of experienced engineers and operators into queryable AI systems. KTBot preserves institutional expertise that would otherwise be lost to attrition.
Energy Theft Detection ML Methodology
Machine learning pipelines for detecting non-technical losses in distribution networks using smart meter data, consumption patterns, and network topology features.
Case Studies
Deployed AI systems and real-world outcomes from SARAL's edge intelligence research.
KTBot Deployment
Deployment of SARAL's knowledge preservation system at an industrial facility, capturing decades of operator expertise into a structured, queryable AI system accessible to new engineers.
Energy Theft Detection on Smart Meters
Deployed ML-based theft detection across a distribution utility's smart meter network, identifying anomalous consumption patterns and reducing non-technical losses.
Related Academy Courses
Edge AI for Energy Systems
Hands-on training in deploying AI models on edge devices for energy monitoring, load forecasting, and anomaly detection in grid infrastructure.
View courseRobotics Inference at the Edge
Building and deploying real-time inference pipelines for robotic systems operating on constrained hardware in industrial environments.
View course