Projects
For further questions on any project, email me to learn more!
LLM Distillation For Financial Reports
Aug 2024 - Dec 2024 | Associated with University of Southern California
- Collaborated with a team of four to enhance accessibility and transparency in financial predictions by leveraging step-by-step distillation techniques for simplifying complex financial reports.
- Benchmarked prominent LLMs, including Meta’s LLaMA 3, Claude 3.5 Haiku, and FinGPT, distilling knowledge into smaller, task-specific models such as GPT-2 and T5.
- Achieved superior performance through distillation (with distilled T5) as compared to non-distilled models such as FinGPT, demonstrating its use as a resource-efficient alternative to fine-tuning for financial tasks.
Skills: Deep Learning · Large Language Models (LLM) · Python (Programming Language)
Emotion Transition And Paraphrasing Using LLMs
Jan 2024 - May 2024 | Associated with University of Southern California
- Collaborated with a team of five members to analyze emotion transition and paraphrasing capabilities of Large Language Models.
- Created curated datasets using existing datasets and fine-tuned LLMs (GPT-2, BART, T5) for paraphrasing and emotion transition with zero-shot, few-shot, and supervised training methods.
- Performed extensive research on six metrics to determine the best-performing model for emotion transition and retaining sentence meaning.
Skills: Large Language Models (LLM) · Natural Language Processing (NLP) · Fine Tuning
Implementation of Abbasi et al. LMBiS-Net
Aug 2023 - Dec 2023 | Associated with University of Southern California
- Implemented first publicly available implementation of LMBiS-Net model deployed in "LMBiS-Net: A Lightweight Multipath Bidirectional Skip Connection based CNN for Retinal Blood Vessel Segmentation" by Abbasi et al.
- Compared performance on 4 older datasets and new datasets to determine applicability.
Skills: Neural Networks · Convolutional Neural Networks (CNN) · Applied Machine Learning
COVID vs. Pneumonia vs. Normal Lung X Rays
Mar 2022 - Jun 2022 | Associated with University of California San Diego
- Utilized image-masks and CNNs to process pixel data and train models to classify a patient’s diagnosis.
- Examined models on hundreds of X-rays to distinguish between healthy, pneumonia, and COVID-positive patients.
Skills: Machine Learning Algorithms · Image Masking · CNNs · Image Classification
Unsupervised ML Techniques on CIFAR-10
Jan 2022 - Mar 2022 | Associated with University of California San Diego
- Compared K-Means Clustering to a combination of PCA and Gaussian Mixed Models to find options with better clustering performance.
- Tested methods on CIFAR-10, a publicly available image database with 60,000 color images of varying categories.
Skills: Principal Component Analysis · Gaussian Mixed Models · Automated Clustering · Unsupervised Learning
Barcode Scanner for Nutritional Information
Mar 2021 - Jun 2021 | Associated with University of California San Diego
- Implemented barcode scanner utilizing webcam to fetch all nutritional data from an online database with thousands of entries.
- Used OpenCV (cv2) for image capture and ZBar (pyzbar) for barcode decoding into the model.
Skills: Python (Programming Language) · Computer Vision · Applied Computer Science
CS:GO Winner Predictions
Mar 2021 - Jun 2021 | Associated with University of California San Diego
- Predicted winner of eSports tournaments by utilizing predictive factors and logistic regression.
- Improved accuracy by choosing the best feature parameters using subset selection and testing models with K-fold CV.
Skills: Logistic Regression · Subset Selection · Cross Validation · Statistical Data Analysis