J·H
N° 00 / Maison J·H S/S 2026 — En vigueur

JazmiaHenry

Principal Research Engineer · AI Research Lead · Thought Leader · Public Speaker · Executive & Commercial Impact

Enter the atelier
N° 01 / 05

On record.

I lead applied AI research from the research question through the system, the evaluation, and the executive decision. My work focuses on post-training, reinforcement learning, agent systems, and scientific evaluation, with a record of building teams and turning machine-learning programs into products used at scale.

I am Principal Research Engineer at Collide, founder of Iso AI, and a former Microsoft applied AI lead and Head of Machine Learning at The Motley Fool. I build technically rigorous systems, set research direction, and present as a thought leader and speaker. I lead high-stakes executive conversations, close executive alignment on technical strategy and commercial opportunities, and connect customer needs to product investment and revenue. At The Motley Fool, ML products I led contributed more than $40M in attributable revenue impact.

Current role
Principal Research Engineer · Collide
Founded
Iso AI · Applied research in agent learning and evaluation
Research
NeurIPS FLLMPT 2026 workshop poster · ISOPro
Education
Stanford HAI · Columbia M.A. · Tulane B.A. · Oxford doctoral studies (DPhil not completed)
N° 02 / 05

Selected work.

Research and platform leadership

RIGGs · Collide

2025 — present · Principal Research Engineer

I lead research and engineering across foundation models, reinforcement learning, agent systems, and evaluation for petroleum engineering. The work spans domain data, post-training, simulation-based learning, knowledge graphs, retrieval, inference, and customer-facing deployment.

Research agenda

  • Post-training — SFT, continued pre-training, DPO, and GRPO for domain reasoning and tool use.
  • Verifiable evaluation — reward dimensions and deterministic checks for reasoning, retrieval, calculations, tool correctness, and long-context behavior.
  • Agent systems — multi-agent workflows connected to domain tools, knowledge graphs, and enterprise data.
  • Research infrastructure — model training, evaluation, serving, GPU planning, and reusable platform components across on-premises and cloud compute.

Executive and customer work

I present technical systems to senior leaders at large energy companies, lead high-stakes conversations about workflow fit and deployment risk, and turn customer evidence into product and research decisions.

  • NeMo
  • Megatron-Core
  • DeepSpeed
  • vLLM
  • SGLang
  • Neo4j
  • Qdrant
  • Ray
Founder and research builder

Iso AI

2024 · Founder and Chief AI Officer

I built ISOPro, an open-source agent evaluation and training framework that uses deterministic verifiers in place of learned reward models. I also built simulation infrastructure for reinforcement-learning research and developed IsoTune, a simulation-grounded fine-tuning approach.

  • ISOPro
  • Agent evaluation
  • RL simulation
  • Patent inventor
Applied research in enterprise systems

Microsoft · Project Bonsai

2022 – 2024 · Lead Applied AI Engineer

I led applied AI and reinforcement-learning work for enterprise customers across industrial and financial services. My work included Azure Plato, a distributed training platform that reduced training time by 80%, and production AI systems developed with teams at BlackRock, Moody's, Shell, PepsiCo, and Abbott.

  • Reinforcement learning
  • Enterprise agents
  • Azure
  • Simulation
Revenue-linked machine learning

The Motley Fool

2020 – 2022 · Head of Machine Learning

I built the ML organization from zero to seven and led recommender systems serving more than one million users. The work increased overall engagement by 33%, increased engagement among high-value users by 42%, and contributed more than $40M in attributable revenue impact.

Look 05 — Provenance

Morgan Stanley

2018 – 2020 · Senior Data Strategist

Built data and quantitative analysis for wealth-management and institutional stakeholders.

N° 03 / 05

Archives.

  1. N° 01

    ISOPro

    Open-source deterministic-verifier framework for agent training and evaluation · NeurIPS FLLMPT 2026 workshop poster

  2. N° 02

    Scaling Laws or the Law of Diminishing Returns

    NeurIPS 2023 Black in AI Workshop · Top 5% poster

  3. N° 03

    Mitigating Bias in Machine Learning

    Book chapter · McGraw-Hill · 2023

  4. N° 04

    MLOps Primer for Policymakers

    CIGI and MIT Press · 2022

  5. N° 05

    Prompting Fairness

    Research on AI fairness and decision-making · arXiv:2402.05786

N° 04 / 05

The runway.

N° 06 · New

The constellation.

A curated map of selected research, repositories, papers, writing and talks. Explore the evidence behind my work.

  • 39selected works charted in the constellation
  • 10research constellations, from RL to evaluation
  • Askthe map anything — it lights the proof
Enter the constellation
N° 05 / 05

Inquiries.

For research leadership, executive briefings, speaking, and enterprise AI partnerships —