Architecture Overview

NataPulse separates data collection, signal construction, reasoning, and presentation. This prevents a raw provider response or model output from becoming a user-facing conclusion without intermediate controls.

1. Data layer

The data layer collects observations from enabled providers. Depending on the domain, these may include social posts, RSS or licensed news items, SEC filings, market candles, market snapshots, blockchain transactions, and block summaries.

Raw observations retain provenance and are processed idempotently so the same source item is not repeatedly treated as new evidence.

2. Signal engine

The signal engine converts observations into structured evidence:

normalization into a common event schema;
entity resolution;
duplicate detection;
source reliability and data-quality assessment;
importance and confidence scoring;
event publication checks;
cluster formation;
narrative derivation;
selected quantitative-signal promotion.

PostgreSQL remains the authoritative structured store. Semantic indexes may accelerate retrieval, but do not replace the source of truth.

3. Reasoning engine

The reasoning layer creates reports, cited answers, impact analysis, and Deep Research. It retrieves only the context appropriate to the task, coordinates specialist agents, exposes uncertainty and contrary evidence, and applies product guardrails before publication.

Relevant completed research can create safe memory lessons for later related investigations.

4. Interface layer

The interface layer exposes curated read models through:

the NataPulse web terminal;
live update channels;
export functions;
selected machine-readable and agent interfaces.

Product interfaces receive explicit, whitelisted fields. Private control-plane scores, secrets, internal traces, model costs, and unpublished content are not part of the public contract.

End-to-end flow

Provider observation
  → validation and provenance
  → normalization and entity resolution
  → deduplication and classification
  → importance, confidence, and publication gates
  → event
  → cluster
  → emerging narrative
  → report / Deep Research / alert / cited answer
  → outcome and memory feedback