FN
Fede's AI & Data Lab

AI & Data Lab

Fede Nolasco

AI & Data Strategy, Architecture and Delivery

Orchestration, Governance & Delivery | Privacy by Design & FinOps

Recent work

What I have been building

Most of my hands-on time this year has gone into five systems that meet the same requirement at different points: a decision about AI should be traceable, and reversible when the evidence changes. Between them they carry 700+ governed task records — completed, dated, and closed on evidence rather than on a status change.

These were not built by hand. The work runs on an agentic delivery model: ten specialist agents — lead, product, architect, reviewer, schema, API, console, visuals, security and docs — under human-in-the-loop (HITL) authority that holds every approval. Work is dispatched as bounded harness units with declared scope and evidence; agents propose, a person decides, and the governed decisions stay under HITL. Model routing is deliberate rather than defaulted: each task goes to the cheapest verified model that can meet its quality, privacy and context requirements, across Sol, Terra and Luna, Gemini, Opus, DeepSeek and GLM. And it sits in a Git-backed project environment — records, ledgers, proposals and decisions as versioned files — with CI gating the changes, from cross-platform and ledger validation on the source pipeline to the governance job on the delivery console.

Delivery governance

No status moves without evidence, and every decision stays reconstructable.

One trustworthy record of what was decided and what was done: status is kept in versioned files in Git and read back through a view that can be rebuilt at any time. I built it because a claim in a chat message, an agent’s memory or a browser view is not a record — only the versioned file counts. Governed changes are stamped so an edit can be traced back, and two separate instances can exchange work without sharing a database.

Architecture change review

Fewer surprises later, because every approval carries its reasoning and evidence.

One proposed change, followed end to end: the business reason behind it, the architecture and engineering evidence it touches, and an explicit human approval or rejection. Nothing advances without its reasoning and its evidence attached.

Diagram catalog

See what a change affects before anyone approves it.

A catalogue of architecture diagrams, each one versioned and searchable. A change adds a new version alongside the old rather than overwriting it, the impact view shows what that change affects, and decisions over the catalogue require a passkey.

Pipeline migration

Runs stop overwriting each other, and every outcome stays on the record.

The article and video pipelines that bring sources in. Runs used to poll a tag and write their status into a spreadsheet column; each run now claims its work from the governed API and reports back its outcome and evidence, so two runs cannot overwrite each other’s state.

Digital assets pipeline

Sources become searchable without paying a model for every item.

The pipeline that turns captured sources into a searchable library. A browser extension and automated feeds bring sources in, each source is identified from the cheapest available signal first and passed to a model only when that is not enough, and what is found is then described and indexed. A measured accuracy test decides whether a change ships, rather than a claim that it improved.

All five are confirmed the same way: a check that can fail, evidence recorded before a status moves, and a person who decides.

About my focus

I work at the intersection of data strategy, AI architecture, and operating-model design, helping organisations turn AI interest into a governed portfolio of decisions.

My focus is on deciding where to invest, what evidence to require, which risks and costs to address, and when an AI capability is ready to scale—or should be redesigned, paused, or stopped.

With more than 20 years of experience in regulated enterprise environments, I connect business priorities, data foundations, governance, and delivery teams. I make trade-offs across value, feasibility, risk, economics, and people explicit, then help the right specialists turn evidence into accountable action.

I treat AI as an operating capability rather than a one-time technology deployment. That means establishing workable decision rights, data and control foundations, human oversight, adoption support, and measures that show whether value is actually being realised.

The model is not the product; the operating system around it determines whether value is realised.

Core Competencies

Enterprise AI Strategy AI Operating Models Strategic Delivery Orchestration Product Vision & Roadmaps Architecture Decision-Making AI Governance & Trust Design Privacy by Design Data Readiness & Quality Enterprise Data Architecture AI Economics, FinOps & ROI Workflow & Agentic System Design AI Evaluation & Quality Assurance Human Oversight & Change Enablement Team Coaching & Capability Building Stakeholder Alignment Hybrid Context Architecture Knowledge Graphs Python & Agentic Coding
Meet Fede in English 1 min intro

This introduction is delivered by an AI avatar, using my own voice. I’m genuinely multilingual and work in English, French, Dutch, and Spanish.

Connect on LinkedIn Fede Nolasco
Key Certificates Credentials in AI, product management, and data science.
Agentic AI Architecture Jun 2026 · Cornell University, eCornell Delivered by Sorin Lerner, Dean; Ayham Boucher, Head of AI Innovations, Cornell Information Technologies (CIT) Enterprise Product Management Fundamentals May 2026 · Microsoft, Coursera Delivered by Marni Baker-Stein, Chief Content Officer The Data Scientist’s Toolbox Feb 2016 · Johns Hopkins University, Coursera Delivered by Jeff Leek, PhD; Roger Peng, PhD; Brian Caffo, PhD — Department of Biostatistics

Video transcript summary: Fede Nolasco introduces his work across enterprise AI strategy, data architecture, agentic systems, structured knowledge graphs, privacy by design, cost-aware delivery, and an open invitation to connect on LinkedIn.

View Projects Get in Touch
20+
Years Experience
EU AI ActGDPR
AI Governance & Privacy by Design
DORANIS2
Operational Resilience & Security
ISO 42010
Enterprise Architecture Discipline