SAURABH GUPTA

Designed and trained a large deep neural model to learn hidden physics of three-phase motors (induction, reluctance, axial flux motor-generator) using simulated and real-world sensor data, enabling advanced anomaly detection and fault prediction.

Developed semi-supervised and supervised training strategies to capture system physics from long temporal basic sensor data, improving model robustness and accuracy.

Engineered a simulation engine to emulate real-world data for induction and reluctance motor generation, optimizing model training for automotive workloads.

Designed and implemented strategies to train edge models using SAC reinforcement learning, enabling deployment on various capacity SOCs (DSP/NPU/GPU).

Managed model distillation, pruning, and quantization processes to port models for real-time inference on edge DSP/NPU, significantly enhancing processing speed.

Architected a real-time streaming engine for ML inference, enabling diagnostic, monitoring, and parameter drift detection using Spark Streaming.

Led the redesign of Matrix Analytics from an on-premise to a multitenant microservice architecture on Kubernetes, significantly improving scalability and efficiency for diverse customer data.

Implemented a real-time streaming aggregation engine using Spark Streaming for generating data, alarms, and anomalies across various dashboards based on user-configured rules.

Developed APIs enabling tenants to configure alarm logic, aggregation rules, and data streams, supporting real-time inference of various pretrained ML models.

Designed and developed platform components conforming to microservice patterns (CQRS, Event Sourcing, Saga), enhancing developer efficiency and consistent design.

Architected common components like Authorization architecture, hierarchical system configurator, and Rule Engines, ensuring robust, observable, and secure microservice operations.

Designed and developed scalable microservice architectures using Spring Boot and FastAPI on Kubernetes, establishing robust platform components for CQRS, event sourcing, and custom integrations.

Architected common enterprise components, including authorization systems, hierarchical configurators, and WebSocket-based notification systems, enhancing system reliability and performance.

Implemented CI/CD pipelines, IAC modules, and test environment strategies, streamlining deployment processes and improving software delivery efficiency.

Developed and deployed stateful streaming applications and real-time model inference engines, processing high-volume data streams for critical business operations.

Designed and implemented APIs for ML models, vector databases, and graph databases, facilitating RAG and Knowledge Graph applications.

Leveraged in-depth expertise in Java, concurrency, reactive programming, and modern design patterns to build high-performance, resilient software solutions.

Led a team of 5 to design and implement an NLP/LLM and Knowledge Graph-based system for recommending Drug Interactions and Adverse Drug Events (ADE) for Pharmacist Drug Utilization Review.

Finetuned BioBert with product databases (MSD manuals, OpenFDA-drug) and adverse drug interaction cases, significantly improving ADE detection accuracy.

Developed a knowledge graph pipeline using Spacy NLP and BioBERT NER to structure patient, prescription, and drug interaction data in Neo4j, enabling comprehensive recommendations.

Designed and led the pipeline for a promotional newsletter recommendation system based on customer purchase history, aiming to increase store visits.

Trained a Deep FM model with sales history and an XGBranker on Spark distributed training, providing the top 20 personalized recommendations and improving promotional effectiveness.