Skip to main content

Pipeline Overview

Every document in Talonic is processed through a deterministic four-phase pipeline. Phase 1 (Resolve) classifies and maps fields. Phase 2 (Agent) performs AI extraction. Phase 3 (Validation) applies confidence gates and quality checks. Phase 4 (Re-read) cross-checks flagged values. Together, these phases produce structured data with per-cell provenance and auditable confidence scores.

Related

Phase 1: Resolve
Document classification and type detection.
Phase 2: Agent
AI field extraction with confidence scoring.
Phase 3: Validation
Confidence gates and quality checks.
Phase 4: Re-read
Cross-checking flagged values.
Four-Phase Pipeline
In-depth pipeline concepts.