Carnegie Mellon University
M.S. in Machine Learning
August 2026 - December 2027
GPA: 4.0 · MLD, LTI
Saee Patil
M.S. in Machine Learning Student Researcher at Carnegie Mellon University
I am an M.S. student in Machine Learning at Carnegie Mellon University, where I am advised by Dr. Akari Asai. My research interests lie in agentic AI and post-training, with a focus on reinforcement learning, reasoning, and evaluation for language models. More broadly, I am interested in building machine learning systems that are capable, reliable, and interpretable.
Research focus
- Agentic AI and language model systems
- Post-training and reinforcement learning
- Reasoning, evaluation, and robustness
- Trustworthy and interpretable machine learning
Education
University of California, Davis
B.S. in Computer Science
September 2022 - June 2026
GPA: 3.91 · Dean's Honor List (7 terms)
Research & Publications
- ICLR 2027Submitted to ICLR 2027
BRIDGE: A Bias Register-Integrated Dataset for Generalized Evaluation of LLM Bias Detection
Patil S., et al.
- NAACL 2027Submitted to NAACL 2027
LLM Capability Is Not a Scalar: Measuring the Shape of Reasoning-Model Behavior with Path Consistency and Sampling Fingerprints
Bonagiri A., Ogus C., Xayasak V., Patil S., et al.
Experience
September 2026 - Present
A2 Lab - Carnegie Mellon University
ML Graduate Research Assistant
Advisor: Dr. Akari Asai
Deep Research Agents: Source Reliability and Post-Training
Developing evaluation benchmarks and reinforcement-learning methods for source-quality reasoning in deep research agents. The work extends citation evaluation beyond factual support to assess claim-specific authority, recency, relevance, and licensing constraints, while using credibility-based rewards and adversarial evidence environments to reduce reliance on outdated or misleading sources.
Deep Research Agents · RL · Post-Training · Citation Evaluation
June 2026 - September 2026
Lawrence Livermore National Laboratory
Machine Learning Graduate Intern
Designed a multi-agent system for CAD-based 3D simulations, combining GEPA prompt optimization with reinforcement learning for finite-element-analysis workflows. Built specialized agents for closed-loop modeling and improved multi-step reasoning and tool use through post-training methods.
GRPO · GEPA · DSPy · Pydantic AI · TRL · HPC
August 2025 - May 2026
Keysight Technologies
Software Engineer (Part-Time)
Fine-tuned CodeLLaMA with QLoRA to automate pytest generation, increasing edge-case coverage by 25-35%, and helped build AWS-hosted test automation for the MARS application, reducing manual test time by 40%.
AWS · Django · MySQL · PyTest · Eggplant · PyTorch · QLoRA
June 2025 - August 2025
Keysight Technologies
Software Engineer Intern (ML & Automation)
Worked on model-assisted automation tooling and internal QA workflows, contributing to the ML pipeline that supported pytest generation and AWS-integrated test automation.
Python · TypeScript · Automation · PyTorch
October 2024 - June 2026
Google Student Developer Club - UC Davis
Tech Director / Software Developer
Led and mentored the technical development track for beginner projects, guided students through system design, web development, ML architectures, and coding issues, and helped build a full-stack portal and backend for TerraAI.
React · MongoDB · Tailwind · Agile
April 2025 - June 2026
PRISM Research Lab - UC Davis
AI/ML Student Researcher
Developed STABLEVAL, a disagreement-aware evaluation framework that models annotator reliability, confusion patterns, and latent item correctness to produce uncertainty-aware scores and more stable AI rankings. Also built a GRPO-driven math reasoning corpus for downstream fine-tuning and benchmarking.
PyTorch · GRPO · Evaluation · Reasoning
April 2025 - January 2026
Davis Automated Reasoning Group - UC Davis
ML Student Researcher
Designed IMPRoViT, a targeted repair strategy for Vision Transformers under adverse-weather distribution shifts, using Captum conductance and top-k Gurobi optimization to identify and repair the pathways causing misclassification.
Captum · ViT · Gurobi · PyTorch
Personal Projects
Pedestrian Intent Classification for AV Systems
Built and evaluated an LSTM sequence model for pedestrian behavior prediction using the JAAD video dataset and YOLOv8, reaching 93% recall on a safety-critical future-intent task.
PyTorch · Scikit-Learn · YOLOv8 · OpenCV
VisionXAI
Developed an interactive application that connects GRAD-CAM explanations and vision-model predictions to a Next.js interface through REST APIs, with NLP-generated clinical context for model outputs.
GRAD-CAM · SHAP · TensorFlow · PyTorch · Supabase · Next.js
Multi-Label News Categorization using BERT
Built a web-scraping and NLP pipeline with BeautifulSoup and Hugging Face, then fine-tuned BERT-base-uncased for multi-label article classification.
BERT · BeautifulSoup · Hugging Face · PyTorch
Breast Tumor Classification
Compared random forests and neural networks for binary breast-tumor classification, documented the full evaluation, and deployed the best-performing model through a Flask interface.
Scikit-Learn · Neural Networks · Flask · Python
HeartHaven
Trained a regression model on heart-disease risk factors and integrated it into a full-stack platform that provides personalized risk predictions and feedback.
Machine Learning · Flask · Python · Full-Stack Development
Con-nect
Built a social-good mobile application that connects returning citizens with professional and educational opportunities, including an LLM-powered interview practice chatbot.
Angular · Ionic · Firebase · OpenAI API