San Francisco, California, United States
Interested in AI agents, neurotechnologies (BCIs), deep learning/RL, and psychedelic research
ServiceNow acquired Moveworks for $2.85B Building the next AI Agent harness for the workplace based on reasoning models.
Built and scaled the Moveworks AI Assistant for employee support. Lead Quality and Evals for our AI Agent system. Focused on AI agent frameworks, building tools/plugins, creating end-to-end evaluation pipelines, implementing customer-facing customization & alignment/steerability tools, and conducting prompt engineering experiments with human & LLM-based annotation (LLM-as-a-judge).
Core contributor to MoveLM, our in-house fine-tuned Large Language Model, specifically for function calling, instruction following, and aligned (safe) outputs. Implemented various synthetic data generation techniques (including Self-Instruct) and lead SFT + iterative DPO training & evaluations/benchmarking.
- Worked on the integration of Natural Language Processing (NLP) models to support Avey’s medical AI assistant designed for self-diagnosis - Researched health insurance claim processing
- Structured and developed a full-stack web app for scaleable data collection of therapists treating NeuraLace's patients - Implemented multiple new features to NeuraLace’s website to prevent spam and optimize NeuraLace’s lead collection process - Created a data reporting and visualization system to improve the team’s data analysis efficacy and efficiency on leads inside Salesforce - Designed experiment and wrote the study proposal for a randomized, controlled dose-optimization study, intended to find an optimal treatment schedule for NeuraLace's AxonTherapy
⁃ Made a sensor network map to simplify complex and large real-world data sets - Preprocessed data using methods for outlier detection/correction, resampling, and missing values imputation (using scikit-learn Python library) - Integrated general methods for evaluating data sets and models, which included defining new dimensionless evaluation metrics - Developed a working pipeline for modeling a predicted feature (using Keras Python library) Worked together with the data analytics research group and engineers from AFRY Schweiz AG, which was the corporate partner for this research.