Angkit Choudhury

Wayne State University

Master’s Degree, Electrical and Computer Engineering

August 1, 2015 – April 1, 2017

National Institute of Technology Silchar

Bachelor’s Degree, Electrical, Electronics and Communications Engineering

January 1, 2009 – January 1, 2013

Siemens

Data Scientist

October 1, 2025 – Present

Austin, Texas, United States · Hybrid

Altair

Data Scientist

October 1, 2017 – September 1, 2025

Austin, Texas, United States · Hybrid

Interactions LLC

Data Scientist

May 1, 2017 – September 1, 2017

Northville,Michigan

Wayne State University

Graduate Research Assistant

September 1, 2016 – April 1, 2017

Detroit Metropolitan Area

Deep language model for solving large anagrams:

February 1, 2016 – Present

Generating meaningful English sentences for a given constraint set of characters : Performance of Deep Language Model: Input:NNNaaaaaaabcccddddeeeeeeeeefffffgghhhiiiiiiiiijkllllllllmmnnnnnoooooooooooooppppprrrrrrrrrrsssssstttttttttuuuuvw<> Output: mr._<unk>_said_the_move_is_expected_to_restructure_decision_to_a_proposal_ and_said_the_move_in_the_rest_of_the_more_market_rose_N_to_N_N_million. Best Anagram-Solver available on internet: Output: Life_Assurance_Ltd_v_Greater_Johannesburg_Transitional_Metropolitan_Council • English PTB data set is tokenized and converted into integer ids. • Log perplexity error on training data is reduced from 770 to 7.8. • Achieved 96% accuracy in small sentences and 92% in long sentences. • Tokenized input and output texts are converted into dense vectors using Word2vec in Tensorflow. • Tensorflow Embedding seq2seq model is used to train the algorithm. • LSTM cell is used for the seq2seq model.

Photo Optical Character Recognition for Autonomous Vehicles

August 1, 2015 – February 1, 2016

• Designed an algorithm to read road signs for autonomous vehicles. • Achieved 99% accuracy on training data, 96% on testing data. • Bumped the accuracy to 98% on testing data using LSTM cell and data parallelism using CUDA toolkit which integrated 4 GPUs( Tesla K40). • The algorithm can perform text recognition, character segmentation and character recognition. • Text recognition is done with sliding window algorithm and trained with a binary classifier. • Character segmentation is trained with binary classifier. • Character recognition is trained with multi class classifier which has 36 classes,26 letters(A-Z)+(0-9) • Programming is done using Python with open source libraries Scipy and Numpy and Tensorflow is used as the building platform.

Embedded System and Sensor fusion: In vehicle control system using CAN protocol

May 1, 2012 – August 1, 2012

• Integrated CAN bus controller, ARM as main control module and LCD display to provide interface. • Developed and implemented a digital driving system to improve driver-vehicle interface. • Used CAN bus(1) to meet real time requirements such as engine speed, wheel speed, throttle pedal locations. • Used CAN bus(2) for locker, windows, mirrors, interior lights feedback controls. • ARM7TDMI-S a 32 bit microprocessor based on RISC principles is integrated with CAN buses and LCD. • In vehicle control system is programmed using Embedded C and debugged with MPLAB XIDE. • A temperature sensor LM 35 integrated with the main module . • A fan and an alarm are programmed to turn on by the main module if the temperature feedback from the sensor crosses 250 degree F.