flan-t5-stance-classification
Parameter-efficient fine-tuning of FLAN-T5 using LoRA for stance classification, with data augmentation, class balancing, and memory-optimized training under limited GPU resources.
ghevde@wisc.edu
university of wisconsin–madison
boston, ma | madison, wi
i'm a bs/ms student in computer science and data science at the university of wisconsin–madison.
i like working on projects at the intersection of machine learning and data systems.
previously, i've been a software engineering intern at fidelity investments, a software engineer at the hawk center for investment analysis, and a teaching and research assistant in the cs department.
i was also president of cardinal trading group, uw–madison's quantitative finance organization.
reach out at ghevde@wisc.edu!
here are a few machine learning and systems projects i've been working on recently. you can see more on github.
Parameter-efficient fine-tuning of FLAN-T5 using LoRA for stance classification, with data augmentation, class balancing, and memory-optimized training under limited GPU resources.
nlp · flan-t5 · lora · text classification
Lightweight implementation of a GPT-style model for character-level text generation, focused on clarity and educational value rather than scale.
language models · transformers · jupyter notebook
Personalized running coach that learns athlete fatigue and recommends daily workouts and routes using Strava data.
time-series · recommender systems · python
Physics-informed neural network that solves the Black-Scholes partial differential equation with boundary conditions and no-arbitrage constraints for option pricing.
pytorch · physics-informed neural networks · pde · numerical methods
Experimental evaluation of Robust Predicate Transfer in DuckDB on the Star Schema Benchmark, measuring runtime, join cardinalities, and memory across scale factors.
python · duckdb · query optimization · data systems
Real-time behavioral firewall that predicts the consequences of a user’s action and intervenes before phishing and social engineering attacks succeed.
javascript · security · real-time systems
programming & scripting: python, java, c++, sql, go, javascript, r, bash, unix/linux
machine learning: pytorch, hugging face transformers, flan-t5, scikit-learn, pandas, numpy, scipy
cloud & devops: aws, google cloud platform (gcp), kubernetes, grpc, docker, jenkins, git, ci/cd, opentelemetry, datadog, grafana, terraform, ansible
data systems & web dev: postgresql, mongodb, dynamodb, bigquery, spark, kafka, fastapi, django, flask, react, node.js, express.js
work experience (view on linkedin)
Graduate Research Assistant, Department of Computer Sciences, University of Wisconsin–Madison
Software Engineering Intern x 2 @ Fidelity Investments, Boston, MA
Software Engineer @ Hawk Center for Investment Analysis, Wisconsin School of Business, Madison, WI
Undergraduate Teaching Assistant, Department of Computer Sciences, University of Wisconsin–Madison
organizations
former president, cardinal trading group