Founding Engineer
June 2025 - Present
Data Engineer
April 2022 - April 2025
Data Engineer, Intern
April 2021 - August 2021
Built experimental pipelines to probe latent world-model representations in proprietary LLMs via structured elicitation tasks and counterfactual prompting. Automated large-scale evaluation runs using Python, PyTorch, and OpenAI APIs, aggregating belief-consistency metrics across 10k+ sampled contexts. Analyzed internal activation trajectories to identify concept clusters and early signs of model self-correction during iterative fine-tuning.
Designed evaluation benchmarks to measure sycophantic behavior in public chat-completion models under varied reinforcement setups. Implemented labeling and aggregation pipelines in Python, Trino, and LangChain to analyze alignment drift across ~500k dialogues. Explored mitigation strategies such as contrastive reward shaping and belief-state calibration, observing consistent reductions in measured sycophancy rates.
Developed adversarial prompting techniques to test and identify vulnerabilities in safety-aligned language models. Designed systematic evaluation frameworks to probe edge cases and failure modes in content filtering and instruction-following systems. Contributed to understanding model robustness and alignment boundaries.
Built a system to process and analyze years of personal journal entries using large language models. Created embeddings and retrieval pipelines to surface patterns, track personal growth, and generate insights across temporal data. Explored privacy-preserving approaches to self-knowledge extraction.
Exploring computational frameworks for implementing synthetic neuromodulatory systems in language models. Investigating how hormone-like signals could modulate LLM behavior and state, drawing inspiration from biological neuromodulation to create more adaptive and context-aware AI systems.
Developed brain-computer interface control system for wheelchair navigation using the Muse EEG headset. Implemented real-time signal processing pipelines to translate neural activity into directional commands, enabling hands-free traversal and improving accessibility for individuals with motor impairments.
Built an interactive system that maps user gestures to high-dimensional latent space coordinates, enabling real-time image generation through diffusion models. Integrated computer vision input (via webcam or gesture sensors) with a generative model pipeline, creating an intuitive and responsive visual output experience. Used Python, PyTorch and OpenCV; deployed models via local APIs and experimented with remote inference setups for scalability.
Built end-to-end surf-forecasting platform aggregating buoy, weather-station, and satellite data (NOAA, Windy API) to predict extreme swell events. Orchestrated automated ingestion and transformation pipelines using Airflow + SQL, storing time-series data in BigQuery for scalable analytics. Developed a React + WebSockets dashboard visualizing live wave trajectories and forecast confidence; integrated Slack/email alerts for major swells.
Designed and trained a deep-learning time-series model predicting hospital bed occupancy across regional facilities using public health and weather data. Built with Python, TensorFlow, and Pandas to reduce prediction error by double-digit percentages.
Free and open-source 3D Slicer Platform enabling users without extensive background to run robust data exploration, visualization, and histopathology correspondence (preprocessing and tissue registration). Build predictive models and visualize results, all in one environment. Created as undergraduate thesis project.
Developed deep learning models to predict gene splicing patterns using genomic sequence data. Applied convolutional neural networks and recurrent architectures to identify splice sites and alternative splicing events.
2023
2023
SPIE Medical Imaging 2023
2023
Columbia Law School Preprint 2024
A weekly gathering for sitting together in public spaces. Featured in CBS News, SFGate, and SF Standard.
Community event bringing people together. Featured in SF Standard.
Various experiments in community building, writing, and making things that probably shouldn't exist but do anyway.
Interested in collaborating on ML systems, data infrastructure, or research projects?
San Francisco, CA