Shakleen Ishfar

University of Rochester

🧠 Recommended and ranked misconceptions associated with incorrect answers to mathematical multiple choice questions.

• Finetuned several large language models (QWEN-2.5, LLaMa-3.1, Phi-3.5) using LoRA and BitsAndBytes to recommend relevant misconceptions for incorrect answer to multiple choice math questions.
• Engineered a 2-stage pipeline to first generate candidate misconceptions using small LLMs and then use large LLMs to rank them, resulting in 0.64 mAP (mean average precision).

University of Rochester Medical Center

🧠 Explored the impact of pharmacological agents on drug-induced migraines by analyzing pulsatility within the glymphatic system.

• Published technical paper detailing an automated computational pipeline for analyzing two-photon micro- scope recordings of mice brain to study drug-induced migraines.
• Leveraged computer vision (ResNET) to mask brain vessels from which diameter was calculated.
• Visualized diameter changes over time using Plotly to identify drug-induced migraine patterns.

University of Rochester

📰 Presented research work which contrasted news from local and international outlets to detect press freedom.

• Finetuned large language model (LLaMa2 7B) to score news articles on the basis of sentiment and stance.
• Utilized BERTopic to cluster news articles into relevant topics for topic wise comparison.
• Designed robust statistical tests to find significant differences between local and international news.

Samsung Research and Development Institute

⌚ Designed smartwatch features to enhance end user experience.

• Implemented three new notification features for the Galaxy Watch 6 series for notification module.
• Improved reliability by increasing test coverage from 30% to 95% and resolving 30+ user-reported bugs.

Intelligent Machines Limited

🏋️ Trained, tested, depolyed, and maintained machine learning models for AI-powered products.

• Trained and deployed object detection model (YOLOv3) to detect store banners for a retail client.
• Engineered social media post analyzer using BERT to gather consumer feedback on marketing campaign.

🧹 Performed data cleaning, processing, and ETL workloads to supply data scientists with clean data.

• Developed image processing pipeline for boosting optical character recognition using OpenCV, Pillow, and NumPy.
• Collected social media data using Scrappy and Selenium.