Thuhin Nandakumar Email & Phone Number

Toronto, Ontario, Canada

London, England, United Kingdom

• Lead the technical development team of 3 students and worked with Project Impactive at University College London, to develop application for those with speech impediments.
• Worked with TensorFlow and Transformers library to create and optimize text prediction for those with speech disabilities. Tested GPT-2, BART and BERT models and trained on conversations scraped form reddit for natural.
• Coordinated project integration between frontend backend API and Machine Learning model, and coordinated responses between project team and advisor

Toronto, Ontario, Canada

• developed frameworks using python and Elasticsearch DB for automating post-silicon verification, data analysis and debugging.
• Overhauled software build system with Docker and Bazel for dependency management and optimized code and to reduce compile times by 60%
• Created frontend graphical user interface using python Textual and CSS, to facilitate testbench management and interfacing, using reactive design principles
• Designed test scripts to automate and conducted ASIC screening system and board screening, reducing the per chip test time by 20% as well as eliminating test errors
• Researched and integrated over the air, software updates using the SW update layer into Yocto Linux build package for IMX8 embedded system evaluation boards, eliminating the requirement for manual updates.
• Measured and optimized ML processor power and performance generated and visualized data for post silicon analytics and Voltage Frequency envelopes to create a generalized power draw prediction machine learning model.

Urbana, Illinois, United States

• Used flutter and Django REST framework to create a backend API with token authentication and permissions, optimized for large scale systems, for box packing algorithm and 5-10minuite video social media platform
• Worked with unity and web sockets to interface between the unity Augmented Reality-core and Django.
• Reached semi-Finals of COZAD New Venture Challenge and presented to potential investors
• Implemented 3-D box packing algorithm heuristic using python, leading to a 20% decrease in volume utilization on average

Toronto, Ontario, Canada

• Benchmarked performance, by evaluating throughput and latency at different payloads of the open libra blockchain and determined optimal payload and block size to maximize data transfer and therefore transactions/second.
• Created a tool to visualize the decentralized voting process and the consensus protocol of libra by modifying the source code using Rust and a python webserver to display the status of a network on compute Canada
• Explored, compared, and contrasted the efficiency and working of different consensus protocols in Hyperledger fabric libra/diem, cosmos and Ethereum 2.0

Toronto, Ontario, Canada

• Co-founded Startup which was selected for the Richmond hill summer company program and $3000 grant reward
• Created for frontend application using react native, achieved responsive application with 0.62ms page speed index
• Integrated system with amazon amplify backend and prepared application for potential scaling

Toronto, Ontario, Canada

- Worked on creating multiple potential neural network architectures for Musical style transfer.- Experimented on the efficacy of generative Adversarial networks compared to traditional Convolutional networks and tuned Hyperparameters to optimize model- Worked using Tensorflow 2.0 and Keras libraries along with librosa for data preprocessing

Toronto, Ontario, Canada

-Part of Computer Vision and Machine Learning Subdivision- Working with a team of 20 on a large-scale project for the International Autonomous Robot Racing Challenge (IARRC) 2020.-Using Open CV for python to detect lane markers and mathplotlib to integrate overlays onto images- Using PyTorch and NumPy and integrating opensource projects to perform signage.

Toronto, Ontario, Canada

-Worked with HDR creator Steve man to develop a semi-autonomous brain-controlled wheelchair - Integrated C++ code for movement controls with input from the Muse Headband.- Worked with Open CV to detect obstacles and alert users

United Arab Emirates