Tapan Sharma

Software Engineer

Munich, Bavaria, Germany8 yrs 9 mos experience
Most Likely To SwitchAI ML Practitioner

Key Highlights

  • Expert in AI-driven video engineering and multimedia processing.
  • Proven success in simulation-to-real transfer for robotics.
  • Strong foundation in deep learning and reinforcement learning.
Stackforce AI infers this person is a Robotics and Multimedia Engineering expert with a strong focus on AI-driven solutions.

Contact

Skills

Core Skills

Reinforcement LearningSimulationDeep LearningGrasp PlanningChatbot DevelopmentData Interface DevelopmentAutomationMultimedia ProcessingVideo Engineering

Other Skills

Assembly LanguageCC++Core JavaData AnalysisFFMPEGGitJavaJavaScriptLinuxMODBUSMatlabMongoDBNegotiationNode.js

About

Multimedia engineering expert. Codecs, transcoding, streaming, AI-focused video engineering, SaaS. Love understanding data and concepts, and how to best utilize them for developing smart solutions. Presently, working as Video Engineer at 3Q GmbH. Past, Multimedia software engineer at Roposo. TU Munich and NIT Hamirpur alumunus. Core skills: Python, C/C++, C#, Java, FFMPEG, Gstreamer, Docker, Keras, Tensorflow, Perceptual Video quality metrics analysis, Deep Learning, Reinforcement Learning & AI-driven engineering.

Experience

3q gmbh

Video Engineer

Feb 2022Present · 4 yrs 1 mo · Munich, Bavaria, Germany

Technical university of munich

2 roles

Master Thesis

Aug 2021Feb 2022 · 6 mos · Munich, Bavaria, Germany

  • Chair of Media Technology
  • PourNet: Robust Robotic Pouring Through Curriculum and curiosity-based Reinforcement Learning
  • Worked on a pure simulation-driven robotic pouring scenario based on Unity3D & Nvidia Flex and curiosity-driven Reinforcement Learning. The agent was trained with domain randomized liquid features using Curriculum Learning. Finally, the pouring policy trained in the simulation was transferred on a real robot, demonstrating a robust simulation-to-real transfer. Observed pouring deviation as low as 2.3 g for dish soap and about 5.5 g for water with many other liquids and pouring containers with different geometries in experiments. This thus demonstrated the robustness of the PourNet to pour different liquids (with different properties) precisely and also use different container geometries.
PythonUnity3DReinforcement LearningSimulation

Research Intern

Apr 2020Oct 2021 · 1 yr 6 mos · Munich, Bavaria, Germany

  • Chair of Media Technology
  • Project: “Robot Grasp Planning with Compliant Gripper Jaws using Deep Learning”
  • 1. Worked with 3D point clouds for stable pose generation and simulation-driven grasp computation.
  • 2. Implemented domain randomization of camera poses for robust dataset generation within the simulation setup.
  • 3. Transferred simulation-data trained grasping policy trained on 8.2 million datapoints on real robot demonstrating sim-to-real transfer of learned skill.
  • 4. Observed a 92% success rate with grasping policy trained using nonplanar surface contact data which was a 10% improvement on traditional approaches on 22 novel objects.
  • Tools: Python, TensorFlow, Scikit-Learn, NumPy, SQL, Pyrender, PyOpenGL, EGL, Docker, GQ-CNN, Ray-based parallelization of pipeline in Python.
PythonTensorFlowDeep LearningGrasp Planning

Itq gmbh

Working Student

Nov 2019Feb 2022 · 2 yrs 3 mos · Greater Munich Metropolitan Area

  • Projects:
  • 1. Worked on the development of data interfaces for a smart batteries project called SmartB4P. The sensory and weather data is collected using MODBUS protocols and pre-processed. Processed data then acts as an input to a reinforcement learning agent being developed by the Fraunhofer institute. The control actions from the RL agent are collected and sent over to batteries over the Modbus link.
  • The communication between the data interface and RL agent is done by Redis PubSub.
  • Language used: Python
  • Libraries: Ray (for multi-processing), pyModbus.
  • 2. Worked on a project called Smart4I, which was an Industrial 4.0 tech demonstrator. The demonstrator was a fully automated customer-oriented product assembly line. A customer could directly request products from the agent and get them assembled with minimum human intervention. I particularly worked on the development of a chatbot as one of the landing points for placing orders. The chatbot was based on the RASA framework in python and used a TensorFlow-based embedding intent classifier. It enabled the chatbot to identify multiple intents from single user input and thus, choose its response accordingly. Also hosted the chatbot on telegram for accessibility.
  • Language used: Python
  • Libraries/Frameworks: RASA
PythonRASAReinforcement LearningChatbot Development

Roposo

Senior Multimedia Processing Software Engineer

Nov 2015Oct 2018 · 2 yrs 11 mos · Gurgaon, India

  • Primarily worked on development of a multi-media processing application.
  • 1. Image Processing: Worked with JAVA, C++, OpenCV and in-house developed libraries to build a scalable application for image cropping, resizing, overlaying, stitching, OCR and Language Estimation, blur detection, object detection, prominent features detection in humans and image compression.
  • 2. Video Processing: Worked with JAVA, C++, Bash, and FFMPEG to build an application for video transcoding (VP9 - WebM, h.265 - HEVC, h.264 - MP4), overlaying, stitching, audio normalisation, speech to text, branding and making videos streamable for end device caching and web streaming.
  • 3. Audio Processing: Worked with JAVA, C++, Bash and FFMPEG to develop an application for audio processing operations like peak normalisation, EBU R128 Normalisation, trimming, channel mixing and transcoding.
  • 4. Other Skills & Technologies: AWS services (EC2, SQS, Lambda, DynamoDB, SNS, S3 and CloudFront), Docker, Maven, SBT, GIT, NPM, Unix, Apache Kafka, Zookeeper (For non-persistent storage along with tools like Kafka)
JavaC++FFMPEGMultimedia ProcessingVideo Engineering

Intellect design arena ltd | a polaris group company

Consultant - Product Engineer

Jun 2015Nov 2015 · 5 mos · Mumbai Metropolitan Region

  • Developer

Society for promotion of electronics culture

Technical Head

Jul 2014May 2015 · 10 mos · Hamirpur, India

  • Organized and managed technical activities of the voluntary student society at the National Institute of Technology, Hamirpur.
  • http://nith.ac.in/spec/about.html

Barc - bhabha atomic research centre

Research Engineering Intern

May 2014Jul 2014 · 2 mos · Mumbai Metropolitan Region

  • Primarily worked on analysis of instrumentation, interfaces and algorithms critical in working of a Nuclear Reactor.

Education

Technical University of Munich

Master of Science - MS — Communications Engineering

Oct 2018Feb 2022

National Institute of Technology, Hamirpur

Bachelor of Technology (B.Tech.) — Electronics & Communication

Jan 2011Jan 2015

DAV PUBLIC SCHOOL, NEW SHIMLA

AISSCE-10

Jan 1999Jan 2010

Stackforce found 100+ more professionals with Reinforcement Learning & Simulation

Explore similar profiles based on matching skills and experience