Arnaldo
Vitaliano
Senior Data Engineer — 18+ years across the data lifecycle. Builds the foundations and the agents that sit on top of them.
in four years at Meta
The shape of the work
A senior data practitioner spanning monetary policy, financial services, and large-scale consumer social sectors. At Meta, has delivered production data products across consumer-social, monetization, compliance, and integrity domains — with a parallel investment in AI-powered analytics tooling. Equally fluent at building the underlying data foundations (pipelines, dimensional models, semantic layers) and the leadership-facing dashboards and AI agents that make those foundations actionable.
Operating principles
Builds with agents as a matter of course — they are the default working surface, not an occasional assist. What follows is the discipline that makes that reliable rather than novel: the context, the evaluation, and the patterns that turn agentic tooling into something a team can depend on.
Agent-first, not agent-assisted
The day’s work happens inside agentic tooling — writing, reviewing, debugging, shipping. Not a lookup of last resort but the surface the work lives on, with the habits and tooling that assumes.
Context is the product
An agent is only as good as the context underneath it. Curating durable, well-tended knowledge — and knowing what to leave out — does more for output quality than any amount of prompt cleverness.
Evaluate like production
Measure against real traffic and real outputs, not synthetic toys. Regression checks, graded results, visibility per case. Ship the evaluation harness before the thing it evaluates.
Compose specialists
A researcher, a writer, a reviewer — narrow agents with clear jobs, orchestrated — beat one omniscient agent every time. Small scopes are easier to trust, easier to debug, easier to replace.
Codify what works
A pattern that worked twice gets a name, a version, and a home. Reusable and parameterized beats one-off prompting — especially for colleagues who want the result without the craft.
Stable contracts at the boundary
Durable interfaces between agents and the tools they reach — warehouses, ticketing, docs, chat. Get the boundary right and the work stays portable as models change underneath it.
Teaching is the multiplier
Learning days, deep dives, hackathons, live demos. Default to teaching, not hoarding — this is how one contributor’s productivity gains become an organisation’s.
Eighteen years, traced
Abuse Detection + Semantic Models (current)
Leads data engineering for an abuse-detection (scraping) modernization initiative and drives semantic-model adoption across the risk org. Rebuilt the underlying session-level data model and modernized the scoring pipeline (~70% memory reduction), isolating it on dedicated capacity and shipping a 6-page ML-pipeline monitor that gives on-call end-to-end visibility in minutes. Co-led adoption of the semantic-model platform behind natural-language analytics agents: bootstrapped domain models across multiple top-level risk domains, authored a reusable scaffolding skill, and shipped the registry dashboard that became the canonical landing surface for semantic models in the org.
Controls Maturity & AI Agents
Co-authored the v1 metrics for an internal controls-maturity measurement framework. Co-developed a family of LLM-powered agents — a general-purpose analytics agent, a controls-maturity agent that turns complex risk data into briefings for senior stakeholders, and a SQL-documentation agent — grounded in retrieval-augmented generation and reusable prompt recipes, used by analysts and engineers across a large risk-and-compliance org. Presented an analytics-agent deep dive at an org-wide AI learning event and pioneered a hackathon approach mapping critical user flows to engagement-weighted risk profiles, informing the following half’s planning.
Privacy & Risk / Accessibility — Senior IC
Moved to the Privacy & Risk org as the data-engineering owner for accessibility compliance measurement. Designed the dimensional model, daily and unified pipelines, and central dashboards leadership uses to track regulatory accessibility compliance — coverage spanning a flagship consumer product (billions of users) and dozens of regulated user-facing flows. Adopted as the source of truth for accessibility compliance reporting in H2 2025 across program management, engineering leadership, and legal / compliance; 100+ landed changes. Took rotating on-call ownership for the privacy-and-risk data team (multi-week shifts, 25–45 tasks each, 0 incident escalations).
Monetization Reliability Scale-up
Scaled the reliability and incident data products: hardened pipelines, built reliability cubes for engineering and management views, ran a large-scale documentation and codemod campaign across the data assets. The underlying schema was templated and reused by an adjacent monetization surface.
Monetization Business Operations — DE
Moved into the monetization business unit’s data engineering team. Launched the org’s first reliability and revenue-fluctuation insights dashboard suite — cube-based pivot models over distributed SQL and real-time analytics — letting engineering teams and stack owners find hotspots and prioritize reliability investment during planning. Established a repeatable playbook that adjacent monetization teams adopted.
Meta — Onboarding / Consumer Social Data (London)
Joined as a Data Engineer in the consumer social product org. Ramped on internal data infrastructure; contributed to creator-facing data products and analytics for the relevance and ranking surfaces.
Banco Central do Brasil — Business Intelligence Manager
Owned the BI surface for Pix, Brazil’s instant-payment scheme — 100M+ users, 1B+ transactions, BRL 500B/month in flows. Built dashboards, graphs, and reports used by the central bank’s payments team and published Statistics of Retail Payments and the Card Market in Brazil. Relocated from Brazil to the UK in mid-2022 to join Meta.
Banco Central do Brasil — Analytics & Research Manager
Led the data work for Brazil’s Financial Citizenship Report 2021, surfacing systemic gaps in financial inclusion and education. Coordinated a nationwide study on risky indebtedness that revealed 5M borrowers exceeding their repayment capacity. Automated the Financial Citizenship Index pipeline, computing results for all 27 Brazilian states from 85M borrowers, 40M low-income individuals, and 23M social-security contributors.
Banco Central do Brasil — Data Engineer
Eleven years at Brazil’s central bank. Delivered the Credit Risk Datamart ETL (Informatica, DB2, Control-M) processing 750M credit operations per month — in continuous production since 2012. Built a Master Data Warehouse integrating 220M citizen records and 1M geographic entities, real-time ETL flows that cut mainframe costs by 20%, a Python address-standardization algorithm across the 220M-record SSN base, and an ML default-probability model covering 85M borrowers. Authored datRprofile, an R package for automated data profiling.
Tools of the trade
Languages & Frameworks
- SQL — advanced
- Python — pipelines, ML/LLM tooling
- Typed server-side languages
- Notebook environments
Infrastructure & Systems
- Distributed SQL query engines
- Large-scale columnar warehouses
- Pipeline orchestrators (Airflow-class)
- Real-time analytics platforms
- Time-series monitoring systems
- BI dashboarding (cube-based)
- Dynamic config management
- Monorepo VCS workflows
- Hermetic build systems
- Container orchestration
AI / LLM Tooling
- Agentic IDEs (Claude Code-class)
- Custom skills + slash commands
- RAG pipelines
- Semantic-model grounding
- Production-driven LLM evals
- MCP server design + integration
- Multi-agent orchestration
- PARA / second-brain workflows
Domains
- Consumer-internet-scale data
- Monetization reliability analytics
- Accessibility compliance measurement
- Abuse / fraud / scraping detection
- Risk & controls measurement
- AI-augmented analytics
- Cross-functional partnership
On stage
How the work happens
Operates equally as a hands-on engineer and as a connector across product, engineering, legal/compliance, and program management.
Strong written communicator. Relies on design docs and one-pagers to align cross-functional stakeholders before building.
Comfortable ramping into new domains every 12–18 months — six product areas in four years.
Bias toward shipping in stacks of small, reviewable changes.
Advocate for AI-augmented developer workflows; early adopter of LLM-driven analytics tooling.