I build the bridge between human problems and technical solutions.
Every organization runs on two layers: the architecture of information, logic and value of its systems, and the human fabric that holds it up. They are optimized together, or neither works.
I come from the sociology of organizations and ended up building the data and AI infrastructure that makes them work. Before touching the technology I read the organization —its actors and networks, the real processes, the information that flows and the incentives that drive it—; then I design, implement and govern those systems for public and private institutions, and I translate between the technical layer and the room where decisions are made. That is what AI adoption and Responsible AI now demand: not policy in a PDF, but systems that work and can be audited.
- cloud / iac
- GCP · Terraform / OpenTofu · secure-by-design IAM · Secret Manager
- backend
- Python · FastAPI · Django · event-driven architecture · HMAC webhooks
- data
- BigQuery · Datastream (CDC) · PostgreSQL · Bronze/Silver/Gold lakehouse
- ai / agents
- Google ADK · agent evaluation (LLM-as-a-judge) · my own MCP servers · OAuth2 · OpenAPI specs · evals
Three planes, one axis
Not three loose services, but three cuts of the same sociotechnical path: build the system, govern it, and get the organization to adopt it. None holds up without the other two.
Platform & engineering
Infrastructure as code on GCP, secure event-driven backends and my own MCP servers in production. Reproducible, auditable systems — not demos.
Responsible AI
Verifiable technical controls —decision logging, lineage, least privilege— mappable to the EU AI Act and NIST AI RMF. Governance as infrastructure, not a PDF.
Adoption & training
From PoC to governed production, and from jargon to decision. I support teams and leadership, deliver training, and translate across the technical, organizational and political.
Already in production
My own MCP servers (Python, OAuth2, OpenAPI specs) exposing APIs and documentation to coding agents, plus an agent evaluation platform built on Google ADK: a deterministic four-state gate and LLM–expert calibration with Cohen's kappa.
12 modules in OpenTofu/Terraform and ~39 reproducible resources across dev/staging/prod, in a regulated fintech.
Backend consolidating ~1.45M events (~780K sales); replaced a legacy batch with ~21h lag by near-real-time ingestion.
Pipeline PostgreSQL → Datastream → BigQuery with zero impact on the transactional database.
Signal-driven AuditLog, lineage and defense in depth; KYC modeled as finite state machines.
Mendoza FuturIA observatory built on the IMIA AI-adoption maturity model — one of the projects I lead.
From regulated fintech, ground transport and application modernization to legislative BI and social programs: systems for public and private organizations where traceability and trust are not optional.
Let's talk about your project
Diagnosis, architecture, implementation or AI governance. Remote, UTC-3, trilingual.