Selected Projects
These proof capsules summarize the three project areas I currently use to show applied AI engineering, data-system architecture, and research-grade public-data work. Evidence is marked conservatively: if a public artifact still needs sanitization or review, it is listed as Evidence pending review instead of linked.
1. Media Monitor
Role signal: Deployed governed news-intelligence and editorial publishing system.
Brief: A governed Argentina-focused system that runs from monitored-source ingestion through immutable run artifacts, versioned contracts, deterministic indexes, AI-assisted briefs/drafts, an explicit human publication gate, publication-aware deployment snapshots, request-time freshness health, and guarded scheduled deployment.
What it proves
- I can design inspectable AI/data workflows across ingestion, enrichment, editorial generation, publication authority, and deployment boundaries.
- I can keep generated drafts separate from durable contract buses and from the explicit human authority required to create public editorial content.
- I can make deployment/currentness observable through deterministic snapshots, public health, freshness policy, and pre/post-deploy guards instead of equating “has a URL” with “live.”
Evidence
- Canonical public outlet.
- Public freshness / deployment health.
- Documentation and architecture map.
- Source repository.
- Explicit human publication gate.
- Publication-aware
site_snapshot.v2contract. - Scheduled guarded public refresh.
- Runtime contract evidence.
- C3 editorial rehearsal / human-gate readiness.
A representative Media Monitor editorial article is deliberately not linked yet. C3 exercised the real promotion and
indexing machinery in isolation without manufacturing human_approved; the first article belongs here only after an
explicit human publication decision.
2. Knowledge / Data Contracts
Role signal: Reliability architecture for AI/knowledge pipelines.
Brief: A contract architecture for making AI and knowledge pipelines inspectable, testable, maintainable, and safer to extend across producers, versioned buses, manifests, observability indexes, and sanctioned consumers.
What it proves
- I can design reliability layers that keep multi-repo AI/data pipelines inspectable and safe to extend.
- I can define boundaries, contracts, observability, versioning, integration governance, and failure containment.
- I can turn knowledge-system architecture into explicit rules rather than hidden coupling between folders and scripts.
Evidence
- Public technical overview.
- Evidence pending review: integration seams page.
- Evidence pending review: run record and observability index contracts.
3. Argentina Poverty Measurement & Atlas
Role signal: End-to-end data engineering, validation, modeling, and public data product.
Brief: A versioned system connecting EPH household surveys, Census microdata, semantic alignment, person-level income modeling, commissioning checks, governed releases, and a public geographic atlas.
What it proves
- I can apply research-grade analytical judgment to public socioeconomic data rather than only moving JSON artifacts.
- I can build reproducible paths from survey microdata to indicators, geospatial analysis, and human-facing outputs.
- I can translate economic research into inspectable maps, charts, tables, and documentation for public audiences.
Evidence
Open Public Data / Poverty & EPH proof capsule