# Platform Architecture

## How we build signal data

From raw public data to structured, enriched intelligence. Six pipeline stages. 35+ sources. One canonical schema.

## The signal pipeline

Six stages from raw public data to delivery-ready intelligence.

### 01 Ingest
We monitor 35+ public data sources continuously: SEC EDGAR, LinkedIn, job boards, patent databases, Glassdoor, Reddit, GitHub, news feeds, earnings call transcripts, and more. New sources added on a rolling basis — went from 25 to 35+ in five weeks.

### 02 Extract
Proprietary AI/LLM models parse unstructured documents — 200-page 10-K filings, podcast transcripts, social posts — into structured signals with typed fields, confidence scores, and entity resolution.

### 03 Normalize
Every signal follows a canonical schema regardless of source: signal_id, signal_type, signal_subtype, detected_at, association, company/contact entity, and a structured data payload. Integrate once — every new signal type just works.

### 04 Enrich
LLM enrichment adds summaries, pain points (with intensity scores), strategic initiatives (with urgency scores), technologies mentioned (with adoption status), and competitor references. Raw data becomes actionable intelligence.

### 05 Resolve
Entity resolution across sources. "Acme AI, Inc." from SEC maps to "acmeai.com" in your CRM. 99%+ domain coverage, 90%+ LinkedIn URL coverage. Cross-source deduplication ensures one clean record per entity.

### 06 Deliver
REST API (sub-200ms), GCS push (daily/weekly), flat file (JSONL, Parquet, CSV), or OEM licensing. Pick the delivery method that fits your infrastructure.

### 35+ Data Sources
### 250M+ Contact Records
### 50M+ Company Profiles
### <200ms API Response Time

## Frequently Asked Questions

### How is this different from intent data providers?
Intent data gives you probability scores based on anonymous browsing behavior from co-ops. We give you verified, timestamped business events with full source attribution. An SEC filing is a fact. A job posting is a fact. Intent is a guess. We do both, but the signal data is what makes us different.

### What does the canonical signal schema look like?
Every signal has: signal_id, signal_type, signal_subtype, detected_at, association (company or contact), entity fields (domain, name, LinkedIn URL), and a structured data payload specific to the signal type. Same schema whether it came from an SEC filing or a Reddit post.

### How fast are new data sources added?
Our pipeline is built for source velocity. We went from 25 to 35+ sources in five weeks. When we add a new source, it flows through the same extract → normalize → enrich → resolve pipeline. Your integration doesn't change.

### What delivery method should I use?
REST API for real-time enrichment and search (sub-200ms). GCS push for batch workflows and warehouse loading. Flat file (JSONL, Parquet, CSV) if you need portable exports. OEM licensing if you're embedding signals into your own product.

### How does entity resolution work across sources?
We resolve entities across all 35+ sources using domain matching, fuzzy name matching, and LinkedIn URL resolution. "Acme AI, Inc." from an SEC filing, "acmeai" from a job board, and "acme-ai.com" from a news article all resolve to the same entity. 99%+ domain coverage.

### Can I use Autobound signals in my own platform?
Yes. OEM licensing lets you embed our signal data directly into your product with white-label support. Current OEM partners include TechTarget, who embedded our API into Priority Engine. Average time-to-market is 4 weeks.

## Start building with signal data
Talk to our team about signal data for your platform.