Hetarth Chopra

why is server rack burning

Lead TA for CS598 - Fundamentals of Data Curation

Part of the Core Detections team, trying to minimize attacks using proprietary ML models

In my research assistantship with the Text Information Management and Analysis (TIMAN) group at the University of Illinois, I will be working under the guidance of Prof. ChengXiang Zhai, a prominent figure in the field of information retrieval and data mining. I will be working on the Community Digital Library Project, especially on adding Large Language Models and RAG implementations on the existing infrastructure.

Working under the guidance of Prof Vikram Adve, on CropWizard (an AI-Based LLM tool to assist farmers in solving queries) and fine-tuning IBM's Ansible Code Generator LLM.

• Acted as the Product Owner of a Language Acceptability Large Language Model by fine-tuning RoBERTa to score a paper's linguistic quality. • Collaborated with the business to develop a data-driven payment model, using Random Forest, yielding a remarkable $1.3 million in savings.

Worked solo on 10+ projects related to Deep Learning, Computer Vision, and Natural Language Processing in the domain of Fraud Detection, User Experience Enhancement and Profit Boosting. Highlights- Enhanced image clarity by fine-tuning DeblurGAN, CNN-based model, in LAB colorspace, increasing Peak Signal-to-Noise Ratio by 72%. • Surpassed Google Translate (Hinglish) with a BLEU of 0.525 by forming a Romanization Python library based on IndicNLP. • Captured 32% more frauds by estimating a product's price using a vector database using MiniLM (Quantized Sentence LLM), K-Nearest Neighbors Semantic Search, and Statistical Z-Score Outlier Detection. • Prototyped a video production engine using After Effects, RabbitMQ, and Apache Kafka, achieving up to 4 video renders per minute. • Fine-tuned a Pre-trained Inception-ResNet and Universal Sentence Encoder MultiModal DL Model to predict with 88% precision of determining an image-name's similarity.

Worked on developing a Machine Learning Algorithm, under Dr.PS Rana, for predicting the concrete compressive strength of concrete by training gradient boosting models on the extracted data from concrete samples.

Relevant Coursework FA23 : STAT 410 - Statistics and Probability II (A) STAT 425 - Statistical Modelling (A+) STAT 432 - Machine Learning, Statistical (A) SPRG 24: CS 444 - Deep Learning in Computer Vision (A+) CS 597 - Independent Study with Dr Vikram Adve (A) STAT 542 - Advanced Statistical Machine Learning (A+) SMR 24: STAT 593 - Internship (with CrowdStrike) (A) FA 24: STAT 431 - Applied Bayesian Analysis (A) STAT 593 - Internship (with Crowdstrike) (A) CS 597 - Independent Study with Dr Vikram Adve (A) SPRG 25: STAT 510 - Mathematical Statistics (B) STAT 437 - Unsupervised Learning (A+)