Guendalina Caldarini — ML Engineer · Researcher

research

Making AI reliable where being wrong is expensive.

Across the systems that reason over knowledge, the agents that act on it, and the regulation that governs both — and proving when their outputs can be trusted.

Temporal knowledge & GraphRAGunder review

Retrieval and multi-hop reasoning that stay correct as the underlying facts change over time.
Agentic AI for financeunder review

Autonomous agents working over financial data — and measuring when they go wrong.
AI safety × regulationunder review

Aligning language models to financial regulation as the EU AI Act's high-risk rules take effect.
Economic data · KG + sentimentunder review

Knowledge graphs and sentiment over central-bank and economic text.

evaluation & auditing — the thread through all of it full publication list on ORCID →

the throughline

Most AI looks impressive in a demo and quietly fails in production. My practice is the opposite bet: build it for real stakes, prove when it can be trusted — and make the machine legible to the people who rely on it. The same instinct runs through the research, the talks, and the teaching.

speaking & keynotes

Three talks, one perspective

Conferences, panels, and university talks across Europe — in French, English, or Italian.

flagship

Being strategic is not selling out

Why the women told to 'just do great work' quietly lose ground on AI access and advancement — and a framework for closing the gap. An argument with data, not a pep talk.

technical

Temporal knowledge graphs and LLMs

From the research: temporal reasoning, knowledge-graph architectures, and LLM reliability in high-stakes financial domains.

organisational

Why AI adoption fails

Most rollouts create awareness but never adoption. What actually changes behaviour: naming resistance, closing literacy gaps, designing for the people who feel most threatened.

about

Built in production, not slides

Seven-plus years building and shipping machine learning in financial-data systems — entity resolution, knowledge graphs, evaluation — and now leading the team behind it. A PhD in deep learning along the way.

An unusual start: linguistics, seven languages, six countries before Paris — where the instinct for making hard ideas legible comes from.

Engineering

Production ML at scale in financial data — the day job, and the reason the rest is grounded.

Research

Knowledge graphs, agentic AI, AI safety, and evaluation — published and under review.

Teaching

MSc Program Director at aivancity — graduate courses, curriculum, and mentoring.

nowSenior ML Engineer · FactSet, Paris

nowMSc Program Director · aivancity

researchKGs · agentic AI · LLM auditing

teachingGraduate teaching & mentoring

contact

Start a conversation

An event, a research collaboration, or a question? Let's talk.

Based in Paris, working across Europe — in French, English, and Italian.