# Varun Bharadwaj > Staff ML Research Engineer @ Amazon AGI | Building Multimodal Reasoning systems (VLMs/LLMs) | ex-Cruise Location: San Francisco Bay Area, United States Profile: https://flows.cv/varunbharadwaj I’m a Staff ML Research Engineer with 12 years of experience in AI and Machine Learning, currently serving as the Technical Lead of Understanding at Amazon AGI working on their Nova models. I specialize in developing multimodal foundation models, reasoning models, and AI agents, with an emphasis on bridging research and reality. In my current role, I partner closely with researchers to define the technical roadmap and vision for our organization. My process - identify the best and most novel ideas across academia and the industry, design clean and controlled experiments rapidly, define evals that predict real-world behavior, and take the most promising ones to production. I strive to deliver innovative and scalable AI solutions that drive impactful outcomes to users and customers. What I work on: • Multimodal foundation models (VLMs/LLMs): images, long video, and documents (OCR, tables, forms) • Agentic multimodal systems: perception, planning, tool use, and learning from feedback • Post-training for reliability: systematic prompting, SFT, and RL for reasoning/agents (RLVR, GRPO-style optimizers) • Evaluation as a first-class product: dynamic benchmarks, slice-based diagnosis, and attribution that explains which intervention moved which failure mode (and why) • Data strategy: synthetic + real data pipelines, mixing/curricula, contamination detection, dedupe, hierarchical clustering, and labeling strategy Recent work significantly moved key multimodal **reasoning** benchmarks (MMLU/GSM8K/HELM, MMMU/MMBench/MM-Vet, Video-MME/Ego4D, OCRBench/OmniDocBench/DocVQA) with strict decontamination and regression gates. Prior to Amazon, I worked at on autonomous systems (Cruise, Rivian), where I led and shipped end-to-end ML systems under real-world constraints: error mining and labeling for continuous learning, long-tail scenario mining, motion prediction modeling, and training/monitoring frameworks. If you’re working on multimodal reasoning, agents, RL, evaluation design modern data curation techniques, I’m more than happy to chat. Specialties: Multimodal Foundation Models, Synthetic data generation, Agentic reasoning, Reinforcement Learning, RLVR, GRPO, Evaluation design, Data mixing/curricula, Long Video Reasoning, Document Understanding, Data-centric improvement methods, Contamination control ## Work Experience ### Staff ML Research Engineer @ Amazon Jan 2024 – Present | Sunnyvale, California, United States ### Staff Software Engineer, ML | Prediction & Planning @ Rivian Jan 2023 – Jan 2024 | Palo Alto, California, United States • Developed novel, advanced Transformer-based models to enhance motion prediction accuracy. • Designed a comprehensive model training, refinement and monitoring framework, with smart actors and metrics, to accelerate planning and prediction algorithm development and model evaluation. • Implemented data extraction and augmentation pipelines, significantly improving input feature quality and diversity for ML models. ### Tech Lead & Senior Software Engineer | AI Platforms @ Cruise Jan 2017 – Jan 2023 | San Francisco Bay Area • Developed Error Mining and Labeling components for Continuous Learning Machine (CLM), a self-supervised system for auto-labeling combined with active data mining, to solve long-tail scenarios. • Enhanced vehicle performance in complex scenarios such as unprotected left turns and pedestrian recognition. • Building distributed systems for HD map creation incorporating real-time and learned priors. ### Senior Software Engineer III | Application Infrastructure @ Cisco Jan 2015 – Jan 2017 | San Francisco Bay Area • Developed and launched an SDN framework for streamlined data center management. • Enhanced application deployment and introspection, significantly improving operational efficiency and visibility. • Created a real-time health monitoring platform for data centers, reducing support calls by 45%. • Built scalable solutions with command-line interfaces and RESTful APIs, accelerating multicloud deployment by over 40%. ### Student Researcher @ Cornell University Jan 2013 – Jan 2014 | Ithaca, New York Area • Conducted research on cross-traffic perturbations, programming packet clustering scenarios using MATLAB, C and Python. • Developed a asynchronous multicasting system in C++, enhancing video viewing experiences for multiple users. Implemented algorithms that increased multicast rate by ~10x. • Collaborated with esteemed professors Prof. Aaron Wagner and Prof. Kevin Tang on research experiments, gaining valuable insights in video processing and network protocols. ### Software Engineer Intern @ Alcatel-Lucent Jan 2011 – Jan 2011 | Bengaluru, Karnataka, India • Authored pipelines that helped customers to resolve broadband network issues independently. • Improved onboarding documentation to facilitate a smoother experience for future interns. ## Education ### Master's degree in Electrical and Computer Engineering Cornell University ### Bachelor of Engineering (B.E.) in Electronics and Communication Engineering Visvesvaraya Technological University ## Contact & Social - LinkedIn: https://linkedin.com/in/varun-bharadwaj --- Source: https://flows.cv/varunbharadwaj JSON Resume: https://flows.cv/varunbharadwaj/resume.json Last updated: 2026-03-22