Valerio Valente, portrait

Valerio Valente.

I build AI systems that survive contact with reality.

If your idea of fun is digging into the messy details instead of waving hands at a slide deck, we are already friends. The work I keep coming back to has one thing in common: the answer matters enough that no one gets to fake it. Regulators have to read it, researchers have to trust it, or factory floors have to act on it. That tends to leave behind the AI-on-stage version and keep the AI-on-disk version, which is the one I find interesting.

When I am not doing that, I am outside pretending I have a balanced life, reading something I cannot put down, or deep into a manga binge I will regret the next morning. I also bake sourdough: controlled fermentation dressed up as a hobby. Same job, really.

Experience

Where I have worked, in reverse order.

  1. 2026 · now

    Data Scientist Aixelerate

    Owning the AI layer of a research platform: agent-driven analysis tools, explainable statistical components, and the product surfaces around them. The job is moving the multi-agent architecture out of dev and into production-grade. In practice, that means deleting the parts that only worked on a good day. The UX gets the same scrutiny as the model.

  2. 2024 · 2026

    Data Scientist InSilicoTrials Technologies

    Designed the company's multi-agent chatbot, including a Model Context Protocol server that gave the LLM real tools instead of hallucinations dressed as ones. Shipped a tabular synthetic-data workflow on a SOTA foundation model, plus the reproducible Azure ML pipelines that kept the research teams from sharing notebooks named `final_v3_real.ipynb`. Placed top 5 in an FDA challenge with a regulatory-compliant RAG and its own evaluation harness.

  3. 2023 · 2024

    Junior Data Scientist Miraitek

    Built a RAG retrieval agent on AWS that cut document-processing time for manufacturing clients, and surfaced enough operational waste through pattern recognition to lift production performance by 7%. Apache Superset + SQL dashboards turned the rest of the guesswork into numbers people could argue about.

Selected Projects

A few things I am proudest of.

  1. 2025

    mcpadapt, upstream contribution

    Upstream contribution to mcpadapt, an open-source bridge between agent frameworks and Model Context Protocol servers. The kind of work that doesn't make a slide deck but quietly fixes friction for everyone downstream of it.

  2. 2024

    FDA challenge, top-5 finish

    Top-5 finish in a precisionFDA challenge with a regulatory-compliant RAG and its own evaluation harness. Built so the answers were traceable enough for regulators to read, not just for users to nod at.

Side Quests

The hobbies are load-bearing.

Reading
Usually mid-book on something I can't put down. The current rotation lives on the reading shelf.
Manga
Occasional deep dives I will regret the next morning, and recommend the morning after that.
Outdoors
Pretending I have a balanced life. Hiking, long walks, anywhere with a view.

Certifications & Awards

Receipts.

  1. Aug 2024

    Deploying Machine Learning Models in Production DeepLearning.AI

    OC06EOAHREQU

  2. Aug 2024

    Introduction to Retrieval Augmented Generation Duke University

    UT2G2PAI109

  3. May 2024

    Machine Learning in Production DeepLearning.AI

    96CN9B7M28XA

  4. Nov 2021

    Huawei Seeds For The Future, top 3.6% Huawei

    Read the article

Hello

Say hi, share a draft, suggest a book.

The fastest reply paths are the social ones below. The form is fine too if a longer note is easier. Either way, I read everything.

Contact form