$ git log --author="pavan"

// My professional journey from mechanical engineering to leading foundation models for robotics.

commit Mar 2026Present

Team Lead — Foundation Models for Robotics

@ Agile Robots SE

// Munich, Germany

Leading the Foundation Models team, defining technical strategy for VLA architectures, pre-training infrastructure, and data pipelines for next-generation robotic agents.

PyTorchVLMsAWS HyperPodDistributed TrainingONNX

commit Aug 2025Mar 2026

Tech Lead — Foundation Models for Intelligent Agents

@ Agile Robots SE

// Munich, Germany

Pioneered a novel VLA architecture attaching action-decoding heads to pre-trained VLMs for zero-shot generalization. Orchestrated end-to-end pre-training on AWS HyperPod across 100+ GPU clusters with >85% utilization. Managed 75TB multimodal dataset with optimized S3 pipelines, reducing data loading latency by 60%.

PyTorchTransformersAWS SageMakerCUDAMixed-Precision

commit Dec 2023Sep 2025

Senior Deep Learning Engineer

@ Agile Robots SE

// Munich, Germany

Engineered a production-grade Robot Transformer policy with causal masking, achieving 85%+ success rates from 5-20 shot demonstrations. Built an internal RAG system using LangChain, reducing retrieval time by 65%. Applied knowledge distillation to shrink models 40% while retaining 98% accuracy.

PyTorchLangChainRAGGYMZeroMQReRun

commit Aug 2021Dec 2023

Software Developer / Deep Learning Engineer

@ Agile Robots SE

// Munich, Germany

Built a modular PyTorch object detection library (YOLO/Faster R-CNN) achieving 45 FPS on edge hardware. Implemented semi-supervised learning with pseudo-labeling, cutting annotation needs by 60%. Overhauled CI/CD pipelines for HPC systems with Docker containerization.

PyTorchYOLOTensorRTONNXDockerGitLab CI/CD

commit Sep 2020Mar 2021

Master Thesis — Deep Learning for Multiple Object Tracking

@ Robert Bosch GmbH

// Hildesheim, Germany

Extended the SHAMANN paradigm for multi-object tracking in crowded scenes. Achieved state-of-the-art stability metrics on MOT17/MOT20 benchmarks with 30% reduced computational complexity through custom attention mechanisms.

PyTorchAttention MechanismsMOTComputer Vision

commit Mar 2020Aug 2020

Deep Learning Intern

@ auvisus GmbH

// Karlsruhe, Germany

Optimized PyTorch models for mobile deployment, achieving 4-7x speedups with ONNX Runtime and TensorRT. Built high-accuracy classifiers from sparse data (<500 samples) using MobileNetV2 transfer learning.

PyTorchONNXTensorRTMobileNetV2Transfer Learning

commit Oct 2019Jan 2020

Teaching Assistant

@ OVGU Magdeburg

// Magdeburg, Germany

Assisted in teaching computer science courses, providing guidance to students on programming assignments and deep learning fundamentals.

PythonTeachingDeep Learning

commit Jul 2016Aug 2017

Executive Engineer

@ Amara Raja Batteries Ltd.

// Tirupati, India

Led production floor optimization initiatives, improving throughput and reducing defect rates through process automation and data-driven decision making.

Process AutomationQuality ControlData Analysis

commit Jul 2015Jul 2016

Graduate Engineering Trainee

@ Amara Raja Batteries Ltd.

// Tirupati, India

Completed rotational training across manufacturing departments, learning production processes and industrial engineering fundamentals.

ManufacturingIndustrial Engineering