Where Pulse fits in the Data-to-AI Value Chain
The operation layer for all five stages, with continuous feedback loops.
Pulse Context connects collection, curation, model training, deployment and impact for live signals, with continuous evaluation and feedback throughout.
The five stages
- 01
Data Collection
Live radio, news and social. Provenance from the moment of capture.
Explore collection → - 02
Data Curation
Normalised signals, common metadata and a source-linked knowledge layer.
Explore curation → - 03
Model Training
Models trained and continuously evaluated against real local signals.
Explore training and evaluation → - 04
Deployment
Grounded answers, agentic governance and playable sources.
Explore deployment → - 05
Impact and Reuse
A citation graph that makes attribution and impact measurement possible.
Explore impact →
Vocabulary and framework adapted from UNDP’s Introducing the Data to AI Value Chain: Promoting Linguistic and Cultural Diversity for AI Diffusion (September 2026), co-published with Current AI.
What happens
at each stage.
01Data Collection
Live collection across broadcast radio, social, news, events and public information. We transcribe radio in real time, with the source playable. Social signals from Instagram, TikTok, YouTube and X arrive via partner surfaces. Every signal keeps its provenance from the moment of capture.
02Data Curation
We normalise, structure, and continuously curate live signals into a knowledge graph. People, places, businesses and events are connected across sources, with common metadata, provenance and the citation graph intact. The same connected knowledge is available to downstream models, agents and humans.
03Model Training and Evaluation
We train and evaluate models against real local signals, so performance reflects the languages, places and contexts the system serves. Continuous evaluation feeds back into the system, with benchmarks grounded in local use rather than a generic English average.
04Deployment
We turn language and cultural capability into something people actually use, with agentic governance built in. The “play the source” feature is the post-deployment audit trail.
05Impact and Reuse
Source provenance and the citation graph make value-return and impact measurement tractable. Communities whose signals Pulse uses can be credited, and impact measured, every time the system is used.
A knowledge graph
of the world around you.
The same place can appear in a radio conversation, a news report, an event listing and a social post. Pulse connects those signals instead of treating them as unrelated documents.
The graph links entities and relationships to the evidence behind them. A changed venue can be connected to the right event; an answer can lead back to the source that reported the change.
That is what makes local knowledge useful across products, not just inside one chatbot.
Resolve the context
Connect references to the people, places, businesses and events they describe.
Keep the relationships
Curate a shared knowledge graph as new signals arrive and circumstances change.
Preserve the evidence
Keep source attribution and the citation graph intact, including the original radio clip.
Under the hood: a context layer for the real world.
What the data-platform world calls a context layer, built end to end on the Data-to-AI Value Chain.
Ingest
Heterogeneous live signals: audio, text, social. Provenance captured at capture time.
Ground
A continuously curated knowledge graph connecting signals, entities and sources. Shared by downstream models, agents and humans.
Answer
Grounded synthesis with the citation graph intact, including the playable source. Built to fail honestly when sources disagree.
REST and streaming APIs. MCP access for AI agents. Webhook ingestion for partner surfaces. Citation-graph export. Audit logs.
One intelligence layer can serve visitor information, hotel services, travel apps and AI agents. The concierge is one demonstration, not the limit of the platform.