Undercurrent
Signal Intelligence Dashboard

Spot what's emerging.
Before it's everywhere.

Undercurrent aggregates signals from news, search, social, and video platforms to surface emerging topics earlier than any single source can alone.

Open Dashboard

What is Undercurrent?

Undercurrent is a multi-source trend detection and narrative intelligence platform. It ingests raw signals from six independent media channels, resolves them into canonical topics, normalizes each source's output to a common scale, and computes a blended score that reflects how fast attention around a topic is growing, not just how much exists.

The result is a dashboard that can surface a topic accelerating across news, search, and social hours or days before it appears on any single-source feed.

Why this matters

Delayed detection costs time

If signals arrive late, teams react late. Decision windows close before action is possible: missed media responses, delayed research pivots, and lost operational windows.

Single-source views are incomplete

No one channel sees everything. A topic can be surging on social while barely registering in news, or vice versa. True breakout requires cross-channel confirmation.

Noise drowns genuine signals

High-volume sources can mask quieter but more meaningful emerging topics. Without normalization and suppression, dominant topics crowd out the ones that actually matter.

Volume alone is not signal

Knowing a topic is mentioned frequently is far less useful than knowing the rate of acceleration, which channels are driving it, and where geographically.

How it works

1

Ingest

Raw signals are collected continuously from six sources: article titles, pageview counts, search interest indices, tweet volumes, and video engagement data.

2

Resolve & Stabilize

Named entities are resolved to canonical topics via Wikidata. Each source's signal is then normalized to a common scale so no single channel dominates the blended score.

3

Score & Surface

A composite score weighs velocity, momentum, source breadth, and geographic spread. Topics with accelerating cross-source attention rise to the top of the dashboard.

Core concepts

Terms used throughout the dashboard and pipeline.

Topic

A canonical named entity (person, organization, event, or concept) resolved to a unique identifier. All source signals for that entity are merged under one topic.

Signal

A measurable indicator of attention from one source at a point in time: for example, a Wikipedia pageview count, GDELT mention frequency, or Google Trends interest score.

Signal Breadth

The number of independent sources supporting a topic. A topic with strong signal across news, search, and social is considered more robustly trending than one spiking on a single channel.

Stabilization

Sources operate at very different scales: pageviews in millions, tweet counts in thousands. Stabilization normalizes each source before blending so no one channel dominates the composite score.

Features

Derived numeric attributes used to assess a topic's trajectory: mention frequency, rate of change (velocity), cross-source agreement, geographic spread, and source breadth.

Label

A classification of whether a topic is trending or trend-worthy. Used during model training to distinguish genuine emerging topics from noise spikes or isolated one-off mentions.

Suppression

A mechanism that down-weights overly dominant or persistently noisy topics so they do not crowd out genuinely emerging signals. Without suppression, high-volume evergreen topics would occupy every top ranking regardless of whether they are actually accelerating.

Raw signals → derived intelligence

Undercurrent transforms unstructured source data into structured, comparable metrics.

Raw input
Derived output
Article titles, headlines
Canonical topic & entity ID
Wikipedia pageview count
Normalized attention signal
Tweet volume, search interest index
Mention count, velocity
Multi-source mention timestamps
Momentum score, trend stage
GKG entity co-occurrence pairs
Topic relationship graph
Cross-source blended signals
Composite score, storyline cluster

Data sources

Six independent channels, each covering a different facet of public attention.

GD

GDELT

Global news and media signal. Monitors thousands of outlets for entity mentions, tone, and geographic distribution in near real time.

WP

Wikipedia

Public attention and information-seeking. Pageview spikes reliably precede or accompany topic breakouts in other channels.

GT

Google Trends

Search demand signal. Reflects active intent (people explicitly searching for a topic), distinct from passive exposure via news or social feeds.

X

X / Twitter

Real-time social reaction. The fastest-moving signal in the blend, useful for detecting initial breakout moments before they propagate to other channels.

YT

YouTube

Video platform momentum. Trending videos indicate a topic has reached a broader audience beyond the news or social early adopter crowd.

BS

Bluesky

Open social chatter via the Jetstream firehose. A real-time, publicly accessible social signal with no API cost or rate-limit constraint.

What the dashboard helps you do

Spot emerging stories early

See which topics are accelerating across channels before they saturate the media cycle.

Compare topic trajectories

Plot multiple topics on the same time axis to understand relative momentum and timing of attention peaks.

See which channels are driving attention

Channel Breakdown and Source Coverage views show whether momentum is news-led, search-led, or social-led.

Explore related entities and storylines

The network graph surfaces co-occurring topics. Narrative Momentum clusters related stories into broader storylines.

Built to extend

Undercurrent is designed around a pluggable adapter architecture. Each data source is an independent module that can be swapped, added, or tuned without touching the core scoring pipeline. Source weights are configurable, allowing down-weighting of noisy channels or amplification of higher-fidelity ones for a given domain.

The same infrastructure that monitors news and social can be redirected to track academic citations, patent filings, earnings call transcripts, or any structured data stream. Undercurrent is domain-agnostic by design. The intelligence layer stays constant; the adapters change.

The Team

MC

Mikel Calderon

Pipeline & Infrastructure Lead

Yahoo!

Full-stack engineer focused on data pipelines, NLP, and real-time analytics. Built the ingestion infrastructure, trend detection engine, and dashboard.

MV

Man Vilailuck

ML Confidence & YouTube Adapter

National Electronics and Computer Technology Center

Focused on ML confidence scoring, signal quality, and YouTube adapter integration.

NL

Noah Lomnitz

SME / X & Google Trends Adapter

Disney

Subject matter expert driving X and Google Trends adapter development and domain requirements.

Get in Touch

Questions, feedback, or partnership inquiries? Reach us at hello@undercurrnt.co.