Machine Learning Deep

Prasoon Kumar

925-***-**** *******.*****@*****.*** Linkedin San Jose, CA

Summary

· CVML Intern working on an LLM-powered recruiting tool aiming to transform the job creation process

· Research experience on methods to improve uncertainty and robustness of the Deep Learning models

· Experience building automotive emission monitors using IoT sensors

· AIML enthusiast inspired by the intersection of technology and human creativity Education

MS in Computer Science Aug ’21- May ’23 University of Texas, San Antonio GPA 3.8/4 B Tech in Computer Science Jul ’17 - May ’21 IEM, Kolkata, India GPA 8.6/4 Professional Experience

MACHINE LEARNING INTERN AUTOMATION ANYWHERE MAR 2024- PRESENT Automation Anywhere, San Jose, CA

· Developing a comprehensive sentence classifier project, customizing models to categorize bots for distinct tasks, transforming images into text, followed by text summarization techniques.

· Hugging Face Lib and PyTorch framework to fine-tune BERT classifier model for sentence classification. MACHINE LEARNING ENGINEER INTERN Sprouts AI OCT 2023- MAR 2024 Sprouts AI, Sillicon Valley, CA

· Created an API and evaluation framework to monitor the features with results indicating 60% reduction in recruiter time spent on job creation, enhancing operational efficiency greatly.

· Working on an innovative job description creation feature that uses - ChatGPT 4, Llama 2, and Mistral

· Specialize in fine-tuning and prompt engineering techniques for performance and functionality optimization GRADUATE RESEARCH ASSISTANT PROJECTS UTSA AUG 2021- OCT 2023 UTSA, San Antonio, Texas

· AI Powered Student Chatbot Aug ’23 - Oct ’23

• Developed a Chatbot that assists students with inquiries on applications, courses, events

• Fine-tuned ChatGPT pre-trained models to craft context-aware responses, seamlessly integrating the chatbot into the univ website, with a feedback monitor system for continuous improvement.

• Chatbot utilized by ~200 students per day, the datasets from which reveal dependable accuracy

· Deep Learning Model Research Aug ’21 - May ’23

• Research study on Uncertainty and Robustness of Deep Learning Models o Employed the Detco Unsupervised method for image classification o Integrated VOS (Virtual Outlier system) for Out-of-Distribution (OOD) detection, to identify and reduce model uncertainty and calibration error.

• Research study on HD(Hyperdimensional) computing model o Performed varied adversarial and non-adversarial attacks to study model’s robustness o Added an inference layer of HD computing with the classical computing o Resulted in 10% accuracy increase and 2.8 times speed of that of classical computing methods for the same data

· Calibration Error Optimization Model Jan ’22 - May ’22

• Created differentiable approximation DECE (Expected Calibration Error) to assess calibration, enabling direct optimization

• Designed a new meta-learning PyTorch framework for tuning hyper parameters aimed at optimizing calibration without necessitating changes to the fundamental model training strategy. Led to 5% accuracy increase.

Prasoon Kumar

RESEARCH INTERN TCS OCT 2020-DEC 2020

TCS, Kolkata, IN

· Built a system called Digi Fleet that utilizes input from diverse IoT sensors installed in cars and gasoline vehicles to predict the carbon emissions of these vehicles.

· Developed a deep neural network model that considers various input factors affecting carbon emissions, including fuel type, vehicle load, weather conditions, engine efficiency, vehicle age.

· Experimented with Recurrent Neural Networks (RNNs) due to the complexity of the dataset. This approach aimed to enhance the model's accuracy in predicting carbon emissions from vehicles by 70%, thereby contributing to environmental sustainability efforts. UNDERGRADUATE RESEARCH ASSISTANT IEM JUL 2020 – MAY 2021 IEM, Kolkata, IN

· Unify Deep Learning Approaches for Vision and Non- Vision Tasks Investigated the role of the state-of-the-art deep neural networks for vision tasks in solving non-vision tasks such as voice and speech recognition and Natural language and vice versa

RESEARCH INTERN Invii.ai MAR 2019-JUL 2019

Sprouts AI, Sillicon Valley, CA

· Worked on a problem of detecting details of human fingers pointing to a book or surface captured by a low- resolution camera in real time. The framework was TensorFlow.

· Contributed to user flow ideation to dataset creation to design and training a Deep Neural Network based Model with 99% accuracy.

Awards / Recognition

Engineering Innovation Contest by Enginx, TCS

· Implemented a smart water management project which was actually deployed by the city of Kolkata India. Used Linear Regression model served via cloud platform utilizing IoT sensors data

· Team of four selected in Top 25. Presentation Link Graduate Courses

Machine Learning, Advance OS, Topics in Deep Learning, Trust in AI, Machine Learning, Data Science, Artificial Intelligence, Large Scale Database Management System, Discrete Math, DS and Algorithms. Skills / Certifications

Prompt Engineering for Developers

Large Language Models by Google

Machine Learning Engineering for Production (MLOps) Specialization Pytorch Tensorflow C, C++ Python Cloud Computing Omniverse Matplotlib Pandas Scikit MySQL Doc MongoDB FastAPI SQL Java JavaScript Interests

In cognizance of mind-body connections indulge in meditation and yoga to unwind.

Contact this candidate