About

Matt Achachlouei

Founder, Rizmi Labs

I started Rizmi Labs to do one thing well: help organizations ship generative AI inside the constraints they actually operate under — regulators, boards, examiners, customers, and the engineers who have to keep the systems running.

For ten years I worked inside a top-4 US bank, in fintech and enterprise architecture, learning what it takes to build software in a regulated environment. For the last four, I've led generative-AI R&D — designing agentic systems for regulatory compliance under SR 11-7, fine-tuning LLMs, building evaluation harnesses, and translating AI work for C-suites and review boards.

My published research is on hallucination detection and class imbalance — the un-glamorous parts of making LLMs reliable. That's the lens I bring to client work: the most important question is rarely "does this demo?" — it is "will this still be working, and defensible, in twelve months?"

Experience

Where the work was learned

  1. Founder · Rizmi Labs

    2025 — Present

    Independent advisory practice on regulated AI: implementation, model risk, governance, and evaluation. Engagements with banks, insurers, and Fortune 500 enterprises.

  2. VP, Generative AI R&D · Top-4 US bank

    2021 — 2025

    Led generative-AI research and development across regulatory compliance and document automation. Designed agentic AI systems aligned to SR 11-7 model risk requirements; built evaluation harnesses, fine-tuning pipelines, and synthetic data programs.

  3. Senior roles, Fintech & Enterprise Architecture · Top-4 US bank

    2015 — 2021

    Worked across fintech product, enterprise architecture (TOGAF, Zachman), and ML platform initiatives. Translated technical strategy for executive decision-making and regulatory review.

Frameworks & methods

What we bring to every engagement

SR 11-7, extended for GenAI

Validation playbooks adapted for non-determinism, prompt drift, jailbreaks, and tool use.

Evaluation as product

Held-out sets, task-grounded metrics, and structured human review — built once, run forever.

Architecture, then code

TOGAF-grade target architectures that engineers can actually build against.

Plain-language risk

Documentation a board, an examiner, and an engineer can each read and act on.

About the name

Rizmi is drawn from Al-Khwarizmi — the 9th-century mathematician whose name gave us the word algorithm. The AI systems we help clients deploy are descendants of that work. We treat them with the same regard for rigor, documentation, and proof.