Michael Allef

Hi, I'm Michael
Allef.

I build AI systems that only say and do what the domain admits: LLM agents with the evaluation to prove it, and machine learning systems that hold up in production.

Applied AI & ML. Hands-on architect at CI&T.

01  About

For 12 years I've built production software, machine learning, and data systems across retail, consumer goods, hospitality, finance, education, and healthcare, the last 9 embedded with enterprise teams. I started in mobile and payments apps, moved into applied data science, and now work mostly on LLM agents.

I also teach what I build, running hands-on workshops on LLM agents and AI coding tools.

I care about pragmatism: AI that actually changes decisions, not demos.

02  Latest

LLM agents. The model interprets what a person means; a deterministic service decides what may happen. Tools are generated from a signed capability catalog, refusals come back as structured notes the agent can act on, and every stated fact is checked against the source system before it reaches a person.

Evaluation. Simulated-user scenarios, hard-bar safety invariants, deterministic graders against the tool-call log, and model triage across providers.

Production ML. Recommender systems, from collaborative filtering and learning-to-rank to two-tower retrieval, and deep-learning demand forecasting.

Research. An independent line of work on latent predictive representations and graph neural networks.

03  Experience

  1. 2025–now

    LLM agents in a regulated enterprise setting: a transactional agent taken from kickoff to a controlled pilot in under three months, with evaluation and grounding built in.

  2. 2023–2024

    Data platforms and recommender systems: a customer 360 built from scratch on Azure and Databricks for a hospitality and entertainment group, and B2B recommendations for a consumer-goods company.

  3. 2018–2023

    Applied data science: deep-learning demand forecasting, recommenders, churn and segmentation for retail, finance, and education.

  4. 2014–2018

    Software engineering and architecture: native Android and iOS apps, highly scalable payment systems (Java, Kotlin, and Go), and tech lead of delivery squads.

04  Elsewhere