Long form

About

I'm a Senior Research Scientist at Meta, where I work on model understanding for LLM agents, attribution, and large-scale ML debugging. My PhD is from the University of Notre Dame's Computer Vision Research Lab (2024, advisor Walter J. Scheirer). Before that, I earned a combined B.S./M.S. in Computer Engineering at RIT, where I also spent four years in Dhireesha Kudithipudi's Nu.AI lab working on low-precision arithmetic, reservoir computing, and neuromorphic-flavored time-series models.

My research line has been consistent: make models you can actually reason about. That cuts two ways — building intrinsically interpretable architectures (prototypical networks, normalizing-flow hybrids), and stress-testing the post hoc explainers most practitioners reach for (SHAP, LIME, perturbation-based attributions) until their failure modes are clear.

Experience

  1. Senior Research Scientist · Meta

    Jul 2024 — Present

    Cambridge, MA / Menlo Park, CA

    • Designed and launched LLM/agent attribution workflows that explain model outputs through prompt segments, retrieved context, tool calls, orchestration, generated responses, and evaluator metrics.
    • Built a production model-understanding platform that moves attribution from notebooks to service APIs — aggregate analysis, eval integrations, third-party model support, and automated prompt/context optimization for agent builders.
    • Connected attribution to intervention: identified low-value or harmful context, guided prompt edits, supported context compression, launched optimization jobs.
    • Designed V2 explanations for a company-wide LLM code-risk model via hierarchical aggregation over code-change hunks; coauthored the accepted AIware 2026 paper.
    • Developed architecture-aware layer/feature attribution for large-scale ranking systems — up to 10× speedup on perturbation-style debugging.
    • Characterized reliability limits of LLM data-influence estimators across noisy / OOD / in-distribution and subdomain settings.
    • Lead maintainer of PyTorch Captum — release engineering, GenAI attribution, type/test health, tutorials. 5.6K+ stars, ~461K monthly PyPI downloads.

  2. Graduate Research Assistant · Notre Dame CVRL

    Aug 2019 — May 2024

    Notre Dame, IN

    • Joint predictive/generative models for interpretable classification (prototypical networks + normalizing flows), state-of-the-art across accuracy, calibration, and density estimation.
    • Demonstrated the infidelity of post hoc explanation methods; built an open-source symbolic framework for studying explanations of arbitrarily complex models.
    • Defended my Ph.D. dissertation in March 2024.

  3. Research Associate Intern · HPE Labs

    May 2023 — Mar 2024

    Milpitas, CA

    • Methods for evaluating and improving natural and adversarial robustness in neural networks.
    • Neural surrogate for a CFD solver (2,000× speedup), combined with online RL to optimize data-center carbon footprint.

  4. Research Intern · MERL

    Jun — Sep 2022

    Cambridge, MA

    • Original research on intrinsically interpretable AI for vision under Dr. Mike Jones.
    • Identified and mitigated a fundamental shortcoming of prototypical-part networks; led to a co-authored patent application and first-author WACV 2024 paper.

  5. Graduate Student Intern · Lawrence Livermore National Lab

    May — Aug 2021

    Livermore, CA

    • Built distributed multi-objective NAS for accuracy/interpretability trade-offs.
    • Scaled Ray-based experiments to 100+ nodes on LLNL HPC.

  6. Research Fellow / GRA · Nu.AI Lab

    Jan 2018 — May 2021

    UTSA & RIT

    • Modeled COVID-19 infectious spread with epidemiologists and demographers; live Texas dashboard for case data and forecasts.
    • Accuracy/energy/latency trade-offs of network compression via low-precision arithmetic and custom hardware.

Open source & service

Tools I work in

Languages
Python · C/C++ · SQL · Bash
ML / DL
PyTorch · Captum · HuggingFace Transformers · scikit-learn · Ray
Research
Feature/layer attribution · data influence · prototypical networks · LLM evaluation · prompt & agent-harness optimization
Infra
Docker · Linux · distributed experiments · service APIs · batch/async workflows · OSS release engineering

Fellowships & awards

Education