Currently a Research Assistant at Math AI Lab (UW) working on reinforcement learning for polynomial synthesis, and a Data Scientist Intern at Fred Hutch developing CAR T-cell therapy models. Passionate about AI research, mathematical optimization, and cutting-edge ML applications. Check out my updated resume for the latest experience and achievements.
I'm a Research Scientist and Machine Learning Engineer with a passion for pushing the boundaries of technology and science. Currently pursuing Applied and Computational Mathematical Sciences with a concentration in Scientific Computing and Numerical Analysis at the University of Washington.
I'm also minoring in Neural Computation and Engineering, focusing on AI applications in Neuroscience. My current research spans reinforcement learning for polynomial synthesis using Graph Neural Networks and PPO algorithms, cancer modeling for CAR T-cell therapy optimization, and cutting-edge machine learning research including deep learning and adversarial ML techniques for understanding complex biological systems and developing robust AI solutions.
When I'm not working on cutting-edge research or building rockets, you can find me playing competitive tennis at UW or dominating board games (especially Monopoly - I haven't lost in years!). I love collaborating with others and exploring new technologies.
September 2025 - Present
Designed a reinforcement learning framework with Proximal Policy Optimization (PPO) and Graph Neural Networks to synthesize efficient arithmetic circuits for polynomials. Achieved ~70% success rate on degree-3 polynomials using curriculum learning, symbolic verification, and shaped reward functions. Currently scaling framework to higher-degree polynomials and integrating Monte Carlo Tree Search (MCTS) for training efficiency.
June - August 2025
Engineered high-performance Python pipelines for large-scale data ingestion, applying vectorization, parallel I/O, and memory-efficient batching concepts from scientific computing to reduce processing latency. Designed and deployed a secure, self-service web application enabling cross-functional teams to configure, trigger, and monitor automated workflows without technical overhead.
August 2024 - Present
Developed mathematical models based on systems of ODEs to simulate CAR T-cell and tumor interactions using parameter sets from B-ALL patient data. Applied linear regression and MSE loss to train models predicting therapy outcomes and assess treatment plan efficacy.
October - December 2024
Developed a reinforcement learning framework using DDPG agents in MATLAB and Simulink to optimize aerodynamic efficiency in axial turbomachinery. Simulated complex physical dynamics with embedded control loops and reward functions targeting lift, drag, and energy efficiency tradeoffs.
September - December 2024
Designed and implemented an advanced real-time eye-tracking system for non-invasive brain-computer-interface to map gaze to on-screen word location. Integrated gaze tracking with EEG headset signals to support a deep learning pipeline aiming to decode cognitive signals into textual output.
May - August 2024
October 2022 - June 2024
Engineered ARES: a GPS-guided rocket recovery system with autonomous drogue deployment logic and fail-safe mechanisms for high-altitude reentry. Built a real-time telemetry pipeline in C++ and Python to acquire, process, and transmit GPS and flight data from onboard microcontrollers.
Recent work plus two in-progress papers
Designed a reinforcement learning framework with PPO and Graph Neural Networks to synthesize efficient arithmetic circuits for polynomials. Achieved ~70% success rate on degree-3 polynomials using curriculum learning and symbolic verification. Currently scaling to higher-degree polynomials and integrating MCTS.
In Progress - Preparing for ICLR/ICML/NeurIPS submissionDeveloped a dual-mode LLM training framework that alternates between low-compute "instruct" and high-compute "thinking" modes. Uses context distillation, LoRA deltas, and MergeKit blending to iteratively boost reasoning on software engineering tasks.
In ProgressMathematical analysis of the Birth-Death model of Oncostreams in Glioma using advanced computational methods. Published research contributing to understanding cancer progression and treatment strategies.
View PaperResearch on undetectable checksum-triggered backdoors in MLPs trained on MNIST, achieving stealth misclassification with advanced cryptographic methods and adversarial ML techniques.
View Research PosterWe investigated whether the model-based planning process of TD-MPC2 could be effectively distilled into a policy network via DAgger, training large-scale fully connected neural networks to imitate planning behavior without performing online trajectory sampling.
View Research PaperInteractive builds with real-world impact
Campus routing app that finds faster, accessibility-friendly paths through buildings and shortcuts to cut walking time. Built with a focus on intuitive UI and real-world commuter feedback.
AI-guided bioinformatics pipeline that proposes FT/TFL1 gene edits to enable new-wood fruiting and robotics-ready orchards. Aggregates genomics data, aligns flowering genes, and uses RL with a neural reward model to score edits.
Systematic empirical study of five representative backdoor defenses across three precision settings (FP32, INT8, INT4) on vision benchmarks. We observe that INT8 quantization reduces detection rates to 0% while leaving attack success rates above 99%, exposing a critical mismatch between defense evaluation and real-world deployment.
Models oncostream cell populations using a birth-death process to analyze glioma aggressiveness and treatment effects (Cytochalasin D). Integrates ex-vivo/in-vivo imaging and Kolmogorov equations to predict population changes under treatment.
Analyzes the Möbius function on the poset of triangular numbers under divisibility; introduces Hasse diagrams, zeta and Möbius matrices, and conjectures about asymptotic behavior, supported by visualizations and experiments.
I love to meet new people and collaborate on exciting projects. Feel free to reach out!
Interested in collaborating on a project or just want to chat about research, rockets, or Monopoly strategies?
Send me a message