🎓 Computer Science + Math (Honors), University of Massachusetts Amherst, Class of 2028 🤖 Focused on Machine Learning, LLMs, and Applied AI Research 🔬 Currently: Data Science & ML Intern @ UMass Dept. of Data Science and AI

I am a rising junior at UMass Amherst majoring in CS and Math. In the past year, I have built projects in machine learning, quantitative finance, and web development. In ML, I worked on mechanistic interpretability in financial time series models for an investment fund, computer vision for sustainability projects tracking ocean waste, and transitional variational autoencoders for synthetic trial generation for rare medicines. In web dev, I worked with the NPO Human Service Forum to enable website development for course recommendations for teachers and students.

📫 Reach me: rudrapatel@umass.edu • LinkedIn

DS4CG — UMass Dept. of Data Science and AI (Jun 2026 – Present)

• Built a reproducible Python + DVC pipeline normalizing 60K+ student literacy records across 30K+ students from 399 raw CSVs.
• Developing decision-tree models to predict student instructional focus from literacy assessment scores, and testing whether prior focus assignments improve recommendation accuracy.

Mechanistic Interpretability on Time Series Models — Minuteman Alternative Investment Fund (Jan – Present)

• Fine-tuned an 805K-param IBM Granite TTM-R2 on options data, cutting MSE by 41% and lifting directional accuracy to 63.7%.
• Ablated internal states to isolate causal pathways in the network's signal prioritization.
• Deployed Ridge linear probes to decode Black-Scholes greeks from hidden layers (peak R² of 0.79 for delta).

OceanEye — Autonomous Underwater Waste Segmentation 🏆 HackUMass 2025 Winner, Best Sustainability Hack Real-time computer vision pipeline for underwater trash detection. Trained & benchmarked YOLOv8 (Nano/Small/Medium) on TrashCan 1.0, hitting 45% Box mAP and 36.2% Mask mAP across 16 classes, running at 30 FPS on consumer hardware. Python YOLOv8 PyTorch OpenCV Ultralytics

Nidaan AI — On-Device Rural Healthcare Assistant Built for the Gemma3n Competition (Google). Fine-tuned Gemma-3n (2B) for offline multilingual symptom triage, field-tested across 2 rural villages in Gujarat with a voice-based interface for 20+ semi-literate users. Gemma-3n Unsloth LangChain (RAG) MERN Google Speech API

ML/AI: Hugging Face Transformers • PyTorch • LangChain • ChromaDB • DVC • RAG • Prompt Engineering • LLM Fine-Tuning Languages: Python • Java • C++ • C • SQL • JavaScript • TypeScript Infra/Tools: Git & GitHub • Google Cloud Run • Flask • Flask-SocketIO • React • JUnit

B.S. Computer Science, Honors College — University of Massachusetts Amherst (GPA: 4.00) Machine Learning Foundations — Cornell University (online certificate, in progress)

_{⚡ Currently exploring: interpretability research and efficient fine-tuning of small language models.}