← All speakers
  • Personal financial-analyst harness ↗

    Galletta and Felippe Felisola Caso jointly built a personal demonstration harness for financial planning, drawing on their experience at Monzo. It was presented in a personal capacity.

    ainotes.us

Bio, Work & Ideas

Francesco Galletta

Conference affiliation: Monzo

Francesco Galletta leads analytics engineering for financial crime at Monzo and co-leads AI strategy for the bank’s data discipline. He has built multiple data teams and works across functions on the next steps for data in machine learning. Outside Monzo, he coaches leaders and builds tools to explore what AI can do. He is based in Madrid. His Sessionize profile describes his work and interests.

Galletta is originally from Venezuela and lived in the UK for about seven years before moving to Madrid. Earlier, he worked as a business intelligence analyst, writing queries and building dashboards. In one instance, a stakeholder spotted a wrong number in a dashboard because it conflicted with knowledge from another team. The experience informs his interest in independent checks on analyses that otherwise look plausible.

Building trust in financial AI

Galletta sees trust as a central challenge when AI helps people make decisions about money. With Monzo colleague Felippe Felisola Caso, he built a personal example of a financial-analyst harness that asks questions to establish shared context, separates user-supplied facts from assumptions, and lets people correct the analysis as it develops. They built it in a personal capacity, drawing on their experience at Monzo.

His own budgeting work helped shape his preference for repeatable calculations. After an agent produced slightly different spending totals across runs, he moved the calculation into a script that could be run again. In the harness, calculations are implemented as code and tested: repeatability provides consistency, while tests help check whether the calculation is correct. He pairs that approach with traceability: people should be able to see where figures and assumptions came from, review consequential decisions, and check results against an independent point of reference. The aim is to keep people accountable while leaving room for the interpretation financial work requires.

Read the topics behind these talks

1 conference talk

Key ideas

Scroll to read ↓

Felippe Felisola Caso and Francesco Galletta explain how Monzo’s existing platform turned a Slack reaction into a reusable agent for coding, incidents and data quality.

  • A simple Slack reaction became useful quickly because Agent Chip could build on existing staging, deployment, monitoring and integration capabilities.
    1:47 ↗
  • An MCP gateway controls access to internal tools and knowledge; sandboxing, defined tools and selected egress constrain remote agent runs.
    5:06 ↗
  • The same agent infrastructure supports incident investigations, squad questions and both reactive and proactive data-quality work.
    10:07 ↗
  • Non-specialists can initiate staged changes, with engineers still responsible for review and deployment.
    12:07 ↗

References