Dhruva Bhat

Built a LangGraph-based multi-agent system for structured/unstructured data to dynamically construct knowledge graphs, enabling scalable GraphRAG pipelines. Developed NL-to-Cypher translation for Neo4j, reducing query effort by 80%. Created a multimodal RAG chatbot (LangChain) across images, docs, and tables with on-prem LLMs, cutting resolution time by 70%. Deployed YOLOX models (92% mAP) and optimized CV pipelines using CUDA, Triton, and Docker, reducing latency by 20%.

• Trained computer vision models using a digital twin of a KUKA robot for pick-and-place operations and successfully deployed the system on the physical robotic arm • Developed a real-time C# API for Tecnomatix Process Simulate, extending simulation capabilities and enabling real-time integration workflows

• Worked on performance Evaluation of PMSM-based solar water pumping systems under partial shading conditions • Implemented Particle Swarm Optimization combined with Gradient Descent and Perturb & Observe MPPT algorithms, resulting in a 0.14% increase in maximum power output

• Developed embedded firmware on the AT89LP5 microcontroller for valve state monitoring and control logic, integrating edge-deployed ML-based anomaly detection for ballast system health monitoring

• Worked on Single Stage PV fed Water Pumping Systems with MPPT Extraction for 3 Phase Induction Motor • Modeled the PV panel from scratch for simulation using single diode model and differential equations • Developed an ANN-based hybrid MPPT algorithm, combining it with incremental conductance to improve power tracking under partial conditions