Graduate Researcher · Robotics & Digital Twin Lab · SJSU
I build reliable embodied and production AI systems—from vision-language navigation to scalable RAG, and Computer Vision platforms.
I am an M.S. Applied Data Science student and research assistant at SJSU, where I study persistent semantic memory, safe agent orchestration, robot learning, and simulation-to-hardware evaluation. Before graduate school, I shipped production AI and distributed systems at Bain & Company and Samsung Research.
Selected work
These are not concept-only demos. Each case study separates the research question, system contract, implementation, evidence, limitations, and next evaluation milestone.
Project archive
Experience
Robotics & Digital Twin Lab · San José State University
Researching vision-language navigation, persistent semantic memory, safe Agentic AI tool orchestration, teleoperation datasets, VLA and behavior-cloning policies, and simulation-to-hardware evaluation on Unitree Go2 and xArm platforms.
Bain & Company · New Delhi
Architected production retrieval and ML systems across 12 workflows, combining dense-and-sparse search, semantic caching, evaluation, and observability. Reduced LLM API cost by 35% and latency by 220 ms; also developed a TensorFlow GNN for general-ledger classification and shipped product workflows around the models.
Samsung R&D Institute India · Bengaluru
Built reliable Java and Spring services, Kafka-based processing, and PySpark/ML workflows for high-volume product data, with operational monitoring and production-oriented failure handling.
About
I’m Dhruv Patel, a graduate researcher in SJSU’s Robotics & Digital Twin Lab. My work asks how language-conditioned agents can perceive a real environment, ground intent in persistent spatial memory, and execute through bounded robotics tools rather than uncontrolled model outputs.
Before graduate school, I built production AI and distributed systems at Bain & Company and Samsung Research. I bring those reliability habits into research: explicit ownership, typed interfaces, observability, reproducible experiments, safety gates, and documented failure modes.
Contact