Automated media ingestion
Sereno automatically reads the main regional sources and recognizes when two stories cover the same incident.
KOHR Atlas · Layer · In production
Sereno. Civic intelligence for Colombia.
Sereno turns news coverage on crime into a living, classified, queryable map. It began in Santa Marta and is built with multi-city architecture to scale to any Colombian municipality.
§ I
What it does
A quantitative reading of the media narrative on security.
Sereno reads local crime news, classifies it, and pins each story to the map. When several outlets cover the same incident, Sereno merges them into a single event.
It's not a reporting system or a copy of official figures. It's a quantitative reading of how media covers public safety — auditable, open, and available to journalists, officials, researchers, and citizens.
§ II
Capabilities
From raw note to canonical, geolocated, queryable event.
Multi-provider AI classification
Each note passes through an LLM that classifies it by category, severity, event date, and relevant entities.
The map as living substrate
Each story is anchored to a point on the map — neighborhood, commune, or coordinate — with a trace of how the location was decided. The map never unmounts: panels (feed, dashboard, investigation) float over it as glass cards, and filters modify what's painted without taking the user out of spatial context.
Specialized extractors
Over each event, Sereno detects and normalizes:
- Individual victims with demographic profile
- Artifacts (weapons, vehicles, objects)
- Judicial milestones (captures, indictments, convictions, escapes)
- Recurring actors and cross-cutting cases
Hex grid + satellite features
Hexagonal grid enriched with GHSL, NDVI, nighttime lights, DEM, texture, OSM roads, and DANE census data. On that substrate, a morphological risk model estimates expected coverage and reveals blind zones.
XGBoost · LightGBM · Neg. BinomialThe /investigate agent
Conversational assistant that surfaces patterns in the narrative and syncs map filters with query context.
Output: the Context view with territorial density, explicit scale, and top stories of the period.
Typed tools · Quantitative evidenceGovernment dashboard
Aggregated view by responsible entity (Police, Prosecutor, Mayor's office) with KPIs, trends, and clearance status. Synthetic reading of the narrative month over month.
Output: the Gov view with KPIs ("104 events · 53 critical"), trend sparkline, victim demographic profile, and map with per-neighborhood event count.
KPIs · Trends · ClearanceMulti-city architecture
Sereno supports any Colombian city — local configuration lives in the database, not in code.
Methodological note
The model describes urban morphology and the bias of coverage — it does not predict real crime. It is a lens on how the media tells security, not a proxy for the facts.
§ ⌖
The product, in use
Three main views — editorial feed, government dashboard, and contextual analysis. See live demo →
/ 01
Government dashboard
/ 02
Contextual analysis
/ 03
Feed
§ III
Who it's for
/ 01
Investigative journalism
Traceability and aggregation over thousands of notes. From the isolated note to the canonical event, with sources preserved.
/ 02
Local authorities
Territorial map with consistent data month over month. KPIs by responsible entity and synthetic dashboard.
/ 03
Academia
Open and reproducible dataset on media coverage and urban morphology. Suitable for research.
/ 04
Citizens
Accessible and geolocated reading of the narrative on security. Open public access.
⌁
Want to see Sereno in action?
A demo in your territory.
[email protected] →
We respond to requests from governments, media, academia, and cooperation.