Atalaya.
A watchtower over Colombian institutional memory. We map, deduplicate and connect what the press already published about corruption cases — we never formulate the accusation, we only make it visible.
Corruption in Colombia is not the property of any one political wing. It is a systemic, deeply interconnected network, embedded in how the country operates — to the point that society has come to normalize it. Atalaya makes that network visible.
Aggregate of the amounts documented in the cases mapped — not a claim of diverted, stolen or indicted money. Atalaya indexes what the press published; it never replaces a judge.
Six stages, in this order.
The sequence is intentional — the UX designer first, then the engineer. The surface and its differentiated value are decided BEFORE the data pipeline.
Cultural framing and opportunity
Name the underlying problem — the loss of public memory in Colombia — and confirm the window of urgency (2026 elections). Define the audience and the expected outcome.
UI sketch and competitive differentiation
Design the visual register and the interaction model. Comparison with PACO and international peers (OCCRP Aleph, OpenCorporates). Result: the force-directed graph as primary navigation.
Features and data flow
What data, where it comes from, how relationships are built. Decision: the Colombian press as sole source; cross-source confrontation as the precision mechanism.
AI classification model
Cascade design: cheap classifier (Haiku) -> robust extractor (Sonnet) with a Pydantic schema; deduplication by news-wire fingerprint.
Technology stack selection
Every layer (ingest -> storage -> processing -> interaction) under operational simplicity and a low cost ceiling: Neon + Pinecone + Next.js.
Visual references and display exploration
Visual references for graph-exploration tools, dossier-style interfaces, editorial typography. The HTML prototype is locked in as the canonical reference.