# Dhruv Patel

**Research Assistant – Robotics & Reinforcement Learning** — San Jose, CA, United States

> Robotics engineer building cross-embodiment learning systems and sim-to-real pipelines for humanoids, quadrupeds, and manipulators; expertise in teleoperation, VLA/RL evaluation, and hardware debugging.

Career Archetype: **Launchpad Trailblazer** (HS-MG) — One of the first Launchpad Trailblazers on Saywise

## Links

- LinkedIn: https://linkedin.com/in/dhruvpatel16
- GitHub: https://github.com/pateldhruv1672
- Portfolio: https://pateldhruv1672.github.io

## Experience

### Research Assistant – Robotics & Reinforcement Learning, San José State University (2026-01 – present)
San Jose, CA · Built a training-to-robot loop for Unitree G1 humanoid, Go2 quadruped, and xArm7 that collects teleoperation and joint-state data, trains PPO/behavior-cloning policies in Isaac Lab, runs simulation gates, and deploys to hardware—cutting setup time from 90 to 30 minutes. Instrumented RGB-D, IMU, and joint-state logs with shared timestamps and offline replay to isolate DDS packet loss, frame calibration drift, and joint-limit violations, reducing debugging turnaround from 45 to 15 minutes.

### President | Applied Intelligence Systems (AIS) Club, Applied Intelligent Systems at SJSU (2025-12 – present)
As President of the AIS Club, I drive initiatives that enhance member engagement and professional growth. • Spearheaded the planning and execution of events and workshops tailored to member interests and industry trends. • Fostered collaboration with faculty and alumni to provide networking opportunities and professional development resources. • Developed strong communication strategies to gather feedback and streamline event planning processes.

### Senior Associate Software Engineer (AI/ML), Bain & Company (2024-04 – 2025-07)
New Delhi, Delhi, India · Fine-tuned Mistral-8B and integrated RAG over Bain's presentation corpus for a slide-generation workflow, automating template selection and data population to deliver 40% productivity gain. Built distributed PySpark/Redshift pipelines processing 1 TB+ daily with schema validation, idempotent execution, and alerting—reducing pipeline rework by 60%. Deployed W&B experiment tracking and CloudWatch observability across 50+ experiments, cutting incident triage from 40 to 20 minutes.

### Research And Development Engineer, Samsung R&D Institute India (2022-06 – 2024-04)
Noida, Uttar Pradesh, India · Led R&D for an instruction-based image-editing system combining LLM instruction parsing with Stable Diffusion, optimizing inference by 250 ms for interactive use. Prototyped real-time English-to-Hindi translation and automated reply generation achieving 450 ms inference for bilingual voice interactions. Profiled and debugged a production LLM image-editing service, reducing p99 latency from 1.2 s to 480 ms through batching, async execution, and resource optimization. Deployed on-device ML models using quantization and knowledge distillation, reducing model size by 60% while preserving 98%+ accuracy.

### Software Engineer Intern, Wipro Limited (2021-06 – 2021-07)
India

### Data Analyst Intern, Edulyt India (2020-06 – 2020-08)

### Research Intern, Sardar Vallabhbhai National Institute of Technology, Surat (2019-12 – 2020-02)
Surat Area, India

## Education

### Master of Science - MS in Applied Data Intelligence, San José State University (2025-08 – 2027-05)
GPA: 3.8/4.0

### Bachelor of Technology - BTech in Electronics and Communication Engineering, Sardar Vallabhbhai National Institute of Technology, Surat (2018-06 – 2022-05)
GPA: 3.7/4.0

## Projects

### The Last Degree – Graduate ROI Intelligence
An end-to-end intelligence platform that bridges the gap between educational investment and job market reality. By unifying university data, live job listings, and layoff trends, the platform provides data-driven career insights via a modern web interface and an AI-powered SQL agent. Technical Overview: Backend & Data Engineering: Orchestrated complex ETL/ELT pipelines using Apache Airflow to ingest and unify disparate university, Adzuna job market, and BLS datasets into Snowflake. Resolved data fragmentation by implementing fuzzy-joins and lookup tables for missing primary keys. AI & Analytics: Architected an LLM-based SQL Agent to allow natural language querying of complex ROI datasets. Integrated an Apache Superset dashboard for enterprise-level visualization of layoff trends and market volatility. Frontend & UI/UX: Developed a high-performance, responsive interface (https://thelastdegree.dev/) to visualize Graduate ROI ratios, active job distributions, and real-time WARN notice trends. Key Achievement: Created a unified data model across 6,400+ institutions and 21k+ active job listings to calculate real-time career "Return on Investment."

### Humanoid Teleoperation & Multimodal Robot Data Pipeline
End-to-end teleoperation capture from motion-capture gloves/suit to Unitree G1 humanoid with synchronized multimodal logging. · Built a Rokoko-to-robot teleoperation pipeline streaming 60 Hz body/hand pose into ROS 2, retargeting motion to the Unitree G1 humanoid and BrainCo Revo2 hand. Collected 120 episodes with 30 Hz RGB video, 60 Hz human pose, robot joint state, and hand commands, maintaining timestamp alignment and dropped samples below 1%. Added one-click calibration, Wi-Fi/DDS health checks, and packet-age alerts to isolate connection faults without full-stack restarts.

### Vision-Language Navigation & Persistent Semantic Memory
VLM-grounded navigation for Unitree Go2 quadruped with semantic map persistence across episodes. · Built a Unitree Go2 vision-language navigation stack that parses natural-language commands, grounds objects in RGB-D observations, associates detections with map coordinates, and sends goals to Nav2—reaching correct landmarks in 50/60 scripted trials. Implemented persistent semantic memory storing object labels and embeddings; on revisits, retrieved previously seen targets 88% of the time versus 63% with only current observations. Classified failure modes and replayed them in Isaac Sim, tuning confidence thresholds and waypoint validation to raise success from 68% to 83%.

### Cross-Embodiment RL & Sim-to-Real Policy Evaluation
Residual-RL adaptation across humanoid, quadruped, and manipulator robots with domain randomization. · Explored COMPASS residual-RL adaptation in Isaac Lab for Unitree G1 humanoid and Go2 quadruped: kept base velocity policy frozen and trained PPO residual corrections for embodiment-specific dynamics, comparing goal success and collision rate over 100 episodes per checkpoint. Applied domain randomization across mass, friction, sensor noise, and command latency for manipulation policies, reducing sim-only failures and improving physical xArm7 transfer success from 62% to 78%.

## Skills

Physical AI, SLAM, Motion Control, Visual SLAM, Robotics, Data Build Tool (DBT), Airflow, Docker, React.js, Snowflake, Retrieval-Augmented Generation (RAG), Systems Design, Natural Language Processing (NLP), Computer Vision, Research and Development (R&D), Large Language Models (LLM), Distributed Systems, Generative AI, Generative Adversarial Networks (GANs), Python, C++, Java, SQL, JavaScript/TypeScript, ROS 2, Nav2, Unitree SDK2, xArm7, NVIDIA Isaac Lab, Isaac Sim, Sim-to-Real, ROS Bags, Reinforcement Learning, Behavior Cloning, Imitation Learning, Vision-Language-Action (VLA), Teleoperation, Motion Retargeting, RGB/RGB-D Cameras, Depth Sensors, LiDAR, IMU, Camera Calibration, OpenCV, PyTorch, CUDA, TensorRT, NVIDIA Jetson, Quantization, W&B, Structured Logging, CI/CD, Linux, ROS 2/DDS Diagnostics, Model Fine-Tuning, Dataset Curation, Multimodal Data Pipelines, Benchmarking

## Credentials

- **Lean Six Sigma for a Sustainable Future** — LinkedIn (issued 2023-08)
- **Mastering Model Context Protocol (MCP)** — LinkedIn (issued 2025-09)
- **Software Architecture Foundations** — LinkedIn (issued 2022-08)
- **Applied Data Science Capstone** — Coursera (issued 2020-05)
- **IBM Data Science Professional Certificate** — Coursera (issued 2020-05)
- **Data Visualization with Python** — Coursera (issued 2020-05)
- **Structuring Machine Learning Projects** — Coursera (issued 2020-04)
- **Improving Deep Neural Networks: Hyperparameter tuning, Regularization and Optimization** — Coursera (issued 2020-04)
- **Machine Learning(Stanford Online)** — Coursera (issued 2020-04)
- **Data Analysis with Python** — Coursera (issued 2020-04)
- **Programming Foundations with JavaScript, HTML and CSS (with Honors)** — Coursera (issued 2020-04)
- **Databases and SQL for Data Science** — Coursera (issued 2020-01)
- **Machine Learning with Python** — Coursera (issued 2020-01)
- **Python for Data Science and AI** — Coursera (issued 2019-12)
- **Python for Data Science and AI** — Coursera (issued 2019-12)
- **Google Cloud Platform Fundamentals: Core Infrastructure** — Coursera (issued 2019-12)
- **Technical Support Fundamentals** — Coursera (issued 2019-12)
- **Neural Networks and Deep Learning** — Coursera (issued 2019-12)
- **Data Science Methodology** — Coursera (issued 2019-12)
- **Data Science Methodology** — IBM (issued 2019-12)

## AI Fluency

- **Claude** — My daily driver for coding and brainstorming ideas (https://claude.ai)

---
Source: https://saywise.com/member7245 (last modified 2026-09-30T20:42:28.268Z)