Nautilus™ · AI Engine · Claude Haiku + Sonnet 4.6

AI That Works While You Sleep

Claude Haiku powers receipt intake, BOL extraction, AP scanning, and SOP onboarding. Five autonomous AI agents run continuously - monitoring compliance, security, infrastructure, platform validation, and institutional memory across the entire operation.

Request a Demo Back to Platform

Live AI Capabilities

Six document intake types. Five autonomous agents. All live in production.

WL-03 · Live

📷

BOL Scan & Extraction

Photo 1-5 images of a vendor's Bill of Lading on mobile. Haiku extracts VendorName, BOL number, trailer number, carrier, receive date, and line items. Pre-fills the receiving form automatically.

Live

🧾

Receipt & Expense Intake

Email receipts to receipts@ or upload directly. Haiku reads the receipt, extracts vendor, amount, date, and GL category. Confidence-gated auto-posting. Duplicate detection across a 30-day window.

WL-04 · Live

📑

AP Invoice Auto-Intake

Drop a PDF or image into the AP Invoices page. Haiku extracts VendorName, invoice number, invoice date, due date, amount, and line items. Returns structured JSON with confidence score.

WL-04 · Live

🛒

Sales Order OCR

Scan a customer's paper purchase order. Haiku extracts CustomerName, PO number, order date, total, and line items - pre-filling the sales order creation form. Source stamped as manual_scan.

Latitude™ · Live

📡

SOP Document Extraction

Upload prospect SOP documents via Sonar™. Haiku extracts operational signals - process types, quality procedures, compliance posture - and maps them to Littoral™ module recommendations.

Manifest™ · Live

📋

Questionnaire Auto-Fill

Sonnet 4.6 reads your SOP library and fills customer questionnaires automatically. 16K/32K token limits, confidence scoring per answer, and human review before anything is sent.

Roz™ · Autonomous Agent

🔍

Roz: Weekly Compliance Monitor

Runs every Monday. Checks training expirations, open NCRs past due date, calibration due dates, and vendor qualification gaps. Generates a narrative compliance report and flags critical findings on the dashboard. No human prompt required.

Triton™ · Autonomous Agent

🛡️

Triton: Security Intelligence

Security intelligence agent. Monitors API access patterns across all tenants, flags anomalous sequences for review, and surfaces failed auth attempts and CORS probes before they become incidents.

Meridian™ · Autonomous Agent

🔬

Meridian: Platform Validation Intelligence

Maps platform coverage against the QMS framework, classifies gaps by severity, and tracks closure across every System Verification Test (SVT). Zero regressions ship to production.

Manta™ · Autonomous Agent

📊

Manta: Infrastructure Cost Tracking

Infrastructure cost intelligence agent. Monitors Azure and DigitalOcean spend in real time, flags unexpected spend against budget, and reports MTD burn across all tenants.

Mnemon™ · Autonomous Agent

🧠

Mnemon: Institutional Memory

Institutional memory agent. Maintains the platform knowledge base, catalogs every platform decision and schema pattern, and helps onboard new team members. 297 entries and growing.

The AI Architecture

How Nautilus™ processes documents

📧

Email Intake Triggers

Azure Logic Apps monitors orders@, ap@, and receipts@ mailboxes. On receipt, the Logic App POSTs the email body and attachments to the Nautilus™ intake endpoint - triggering Haiku extraction immediately.

📊

Confidence Scoring

Every extracted field carries a confidence score (0-1.0). Fields below 0.70 are flagged amber in the review UI. No data is posted to the database without a human acknowledging low-confidence fields.

🔍

Duplicate Detection

Expense intake checks for duplicates within a 30-day rolling window on vendor name + amount + date. Suspected duplicates are flagged with PossibleDuplicateOf reference - held for review before posting.

📝

Audit Log on Every Extraction

Every Haiku invocation is logged: BOL_SCAN_ATTEMPTED, SO_INTAKE_SCANNED, AP_INTAKE_SCANNED. The audit log captures the raw JSON response, confidence, and whether the result was accepted or rejected.

🧩

Anthropic Beta PDF Support

PDF documents use Anthropic's pdfs-2024-09-25 beta document block - direct PDF reading without conversion. Images use standard vision blocks. Both paths normalize to the same structured output schema.

🔒

Within Your Compliance Boundary

All AI processing runs server-side via the Littoral™ API - your document data does not leave through the browser. API key secured in environment variables, never exposed to the frontend.

AI Compliance Agents

Running Every Monday

Meet Roz™: Your Autonomous Compliance Monitor

Roz is an AI agent powered by Claude Sonnet 4.6. She runs every Monday, scans your entire operation, writes a narrative report, and flags issues on the dashboard before the work week starts.

The Weekly Checks

Training expirations approaching or already past due

Open NCRs with no action past their due date

Calibration due dates coming up or missed

Vendor Qualification Program gaps - approvals lapsing or incomplete

Vendors accumulating 2+ open SCARs (repeat-offender detection)

CAPAs open longer than 90 days. Escalates to critical at 180 days.

✍️

Written Narrative Report

Roz doesn't just return rows of data. She writes a paragraph-form compliance narrative: what she found, why it matters, and what should happen next. Plain English, not JSON.

📊

Dashboard Integration

Critical unresolved findings appear as a red card on the management dashboard. Count goes to zero when findings are resolved. No report means a clean week, acknowledged with one click.

⚡

Run On-Demand

Administrators can trigger Roz manually via the /admin/roz page. Response fires immediately, scan runs in the background. New report appears within 30 seconds for most production databases.

💬

Example Roz Narrative (real output)

"This week's scan identified 3 high-severity findings. Two raw material profiles (Maltodextrin 20DE and Calcium Carbonate USP) have been revised since their last approval but are not linked to an active MOC record, which creates a gap in your change control audit trail. Additionally, Pacific Ingredients has 2 open SCARs with no corrective action documented in over 45 days. Recommend escalating supplier contact before the next PO release."

AI with a compliance-first design

Nautilus™ is designed for regulated manufacturing environments. Every AI action is audit-logged with user identity, timestamp, and raw model output. No AI result is persisted to the database without human review and explicit approval. Confidence scores are surfaced - never hidden. The system's posture is always: AI suggests, human confirms.

See Nautilus™ in action: documents and Roz

Bring a receipt, a BOL, or a vendor invoice to the demo. We'll run Haiku extraction live and trigger a Roz scan so you can see the compliance narrative in real time.

Request a Demo Back to Platform