CV
Education
Harvard University, MS in Data Science · 2022
Wesleyan University, BA in Mathematics and Computer Science, Phi Beta Kappa
· 2018
Publications
Style Outweighs Substance: Failure Modes of LLM Judges in Alignment Benchmarking
Feuer, Goldblum, Datta, Nambiar, Besaleli, Dooley, Cembalest, Dickerson. ICLR 2025. arXiv
Reckoning with the Disagreement Problem: Explanation Consensus as a Training Objective
Schwarzschild, Cembalest, Rao, Hines, Dickerson. AIES 2023. arXiv
Tensions Between the Proxies of Human Values in AI
Datta, Nissani, Cembalest, Khanna, Massa, Dickerson. SaTML 2023. arXiv
Work Experience
AI Labs Engineer · ExodusPoint Capital Management · 2026–present
Building evals, memory, personalization, monitoring, and self-improvement loops for AI systems, alongside research.
Developer Advocate · Nomic AI · 2024–2025
Owned the docs and SDKs that tens of thousands of developers used to build with Nomic's embedding models, and benchmarked the models against frontier labs to demonstrate their performance and cost-effectiveness.
Researcher · Arthur AI · 2022–2024
Researched and built LLM evaluation and security systems, benchmarking LLMs to guide which models Arthur's product could rely on.
Math & Computer Science Teacher · African Leadership Academy · 2018–2020
Taught 11th & 12th grade math, statistics, and computer science. Worked with students as part of the House program to conduct games, competitions, and bonding events for students, faculty, and staff.
Projects
2025
Published research analyzing LLM-judge failure modes for AI alignment. Developed SOS-BENCH benchmark to provide more reliable measurements of model behavior.
Contributed a guide to the official UMAP documentation demonstrating neural network training visualization using Nomic Atlas.
Owner of the technical guides, developer SDKs, and explanatory videos for the Nomic AI Platform. I was responsible for explaining Nomic's embedding models and overall product capabilities into a comprehensive library.
Delivered a talk at the Open Data Science Conference (May 2025) detailing Nomic's pre-training and post-training methods for embedding models.
2024
Conducted research on the end-to-end lifecycle for LLM experimentation and development, including model selection, prompting, tool use, inference acceleration, structured outputs, and more.
Code & walkthrough for training graph neural networks using PyTorch Geometric and visualizing learned embeddings during training.
Presented "The Ultimate Guide to LLM Experimentation and Development" at the Open Data Science Conference.
2023
We formulated a new training objective to improve the consistency of neural network explanations. Presented at the 2023 AAAI/ACM Conference on AI, Ethics, and Society.
Published experiments evaluating large language models, covering methodologies, benchmarking, and implementation strategies.
Presented "Reckoning with the Disagreement Problem: Post-Hoc Explanation Agreement as a Training Objective" at the Open Data Science Conference.
2022
Position paper on the tradeoffs between fairness, transparency, and privacy in AI. Presented at the 2023 SaTML Conference on Secure and Trustworthy Machine Learning (IEEE).
Article explaining Shapley values and calculating their residuals to measure explainability limitations. Published on TowardsDataScience.
Developed interactive web apps to help make AP Statistics lessons more engaging and relevant to students, covering distributional sampling, hypothesis testing, and the Central Limit Theorem.
Final project for Harvard CS171 Data Visualization. A playable synthesizer in Unity where each instrument is visualized as a 3D object deforming based on spectral analysis.
2021
Developed a geospatial model to identify likely locations of lead pipes in Flint, Michigan and prioritize digging efforts in the most urgent areas.
2020
Final project for Harvard CS107 Systems Development. A Python package for automatic differentiation with forward and reverse mode, plus second-order Hessian computation via Edge Pushing.