Gokul Sunilkumar

Experience

• 

  NASA - National Aeronautics and Space Administration

  Oct 2015 - Oct 2015

  Student Trainee

  • Devised the prototype of a landing rover that carried a payload from a height of 18 feet and land safely.• Underwent zero-gravity training for a period of three days along with whole-body shuffles and other activities.

• 

  DL E&C Co., Ltd

  Nov 2019 - Dec 2019

  Winter Intern : Process Control System and Plant Information Management System

  • Engineered a customised Process Control System and Plant Information Management System (PIMS) at Jacob’s Zate at Daelim Industries’ design headquarters office in Seoul, South Korea. • Designed a challenging PCS that monitors industrial processes to make sure they don't vary from pre-set limits by starting corrective action if they find unwanted deviations or abnormalities. • Formulated a closed-control loop, a controller and a sophisticated regulator which is still being used in their factory to date• Boosted efficiency as the computerisation increased the ability to share data between departments. Show less

• 

  Microsoft

  Apr 2020 - Jun 2020

  Machine Learning Intern

  • Developed Class Activation Maps(CAMs) for pre-developed industry-standard Convolutional Neural Networks(CNNs) and applied GRAD–CAM, GRAD-CAM++, Visual Question Answer Based Techniques(VIQAR), Causal Analysis, etc to industry-specific problems in this field as they provide better localization of objects as well as explain occurrences of multiple objects of a class in a single image. • Deducted a paradigm that helps in efficiently providing better visual representations of CNN model predictions helping us reassure the fact that the model created is trustworthy in all aspects and every single decision of the model is accounted for and has an associated reason. Show less

• 

  International Institute of Information Technology Hyderabad (IIITH)

  Jun 2020 - Aug 2020

  Summer Research Fellow - Indian Academy of Sciences

  • I managed the development of neural network interpretability through a research project on Explainable AI for Sequence Processing Neural Networks.• A 6-node Reber Grammar framework was created to let neural networks simulate language and process grammar with ease.• To improve behavioral insights, agglomerative clustering was used to investigate learning patterns in Elman, LSTM, and Echo State networks.• By tuning reservoir matrix hyperparameters, Echo State Networks (ESNs) yielded the best accuracy and outstanding outcomes.• Through the extraction of useful insights from the identified clusters, significantly contributed to the advancement of AI interpretability. Show less

• 

  Robert Bosch Centre for Data Science and Artificial Intelligence (RBCDSAI)

  Jun 2021 - Dec 2021

  Research Intern

  • I oversaw the development of a transformer decoder with dual attention that integrates split source embeddings after encoding from two different sources found via topic modeling and uses a curiosity-driven question generating mechanism for reference knowledge.• I used a two-stage approach: first, using the factual context as a guide, reference information was established. Next, questions were developed using this reference knowledge.• I worked cooperatively with a group of five analysts to accomplish the project's goals, outperforming the best model available at the time by 0.015 and earning a final BLEU-4 score of 25.42. Show less

• 

  GE Aerospace

  May 2023 - Aug 2023

  AI/Machine Learning Technical Intern
• 

  JPMorganChase

  Sept 2023 - Jan 2024

  Data Scientist

  • I examined and discovered errors in the existing techniques for GPTNeo-X 20B language model factual memory updates that are done directly.• I oversaw the improvement of the creditworthiness and loan acceptance algorithms, improving the language model alterations' specificity and efficacy.• I tested changes to an improved algorithm on the zsRE and CounterFact platforms in order to ensure specificity and avoid any unintended consequences.• I inferred and demonstrated that the properties of the subject, object, and relation in the prompt determine where the edit is located in the base transformer model.•I participated in continuing data science and machine learning research, actively influencing the development of credit evaluation techniques. Show less

• 

  Columbia University

  Dec 2023 - now

  Staff Associate Data Scientist

  Department of Earth & Environmental Engineering• Analysed the effects of different levels of implementation rates for Non-Sewer Sanitation Systems• Automated the process of combining BioWin simulated summary reports in order to create faster visualisation dashboards