AI Autonomous Supply Chain Decision Automation

The Xakot × Opus AI solution automates document-driven supply-chain decisioning through a reusable Intake → Understand → Decide → Review → Deliver pipeline. It handles real-world messiness—PDFs, images, emails, JSON, inconsistent fields, missing data, and mixed compliance checks—demonstrated through a food-retail scenario where a supermarket validates deliveries from a supplier. Intake & Processing: The system ingests multi-format documents (JSON, PDFs via OCR) and enriches them through parallel external sources: OpenFoodFacts for product metadata, OSRM for ETA/distance, a mocked ERP pricing API, and a temperature-monitoring API. All inputs are merged into a canonical schema through a normalization layer that reconciles quantities, units, pricing, safety data, and logistics attributes. Hybrid Decisioning: Deterministic rules (SKU allowlists, halal certification, temperature thresholds, currency checks, quantity mismatches) run alongside an Opus Agent that interprets nuance, assigns confidence scores, and recommends approval, partial acceptance, or rejection. Parallel branches reduce latency and improve throughput. Quality & Safety Review: Two review layers ensure safety. • Agentic Review enforces non-negotiable policies such as halal certification and food safety. • Human Review handles low-confidence or high-impact cases, allowing overrides and notes. Provenance & Audit: Every step produces a detailed, hash-linked audit artifact containing inputs, extracted fields, rule violations, AI rationale, reviews, timestamps, external API sources, and the final decision. This guarantees transparency and traceability. Delivery & Reporting: Results are exported to Google Sheets, email, or JSON stores, with metrics such as trace_id, risk level, and violations. The solution also supports batching, chunking for long OCR text, timeout mitigation, schema fallback logic, and an optional operator console for monitoring and retries.