Vanta-Pilot: Architecture & Setup Guide

This guide details the technical foundation, configuration options, and deployment requirements for the Vanta-Pilot AI engine across Edge Home Solutions.

---

1. High-Level Architecture

Vanta-Pilot utilizes a hybrid intelligence pipeline combining Google Gemini LLM reasoning with ERPNext native relational telemetry:

graph TD
    User([ERPNext Desk User]) -->|Click 'Ask Vanta-Pilot'| API[scripts.vanta_pilot.api.analyze_document]
    API --> Router{Agent Router}
    Router -->|Lead / Quotation| CRM[CRMAgent]
    Router -->|Material Request / Supplier Quote| Proc[ProcurementAgent]
    Router -->|Stock / Workshop| Inv[InventoryAgent]
    Router -->|Sales Invoice / Milestone| Fin[FinanceAgent]
    
    CRM & Proc & Inv & Fin --> LLM[GeminiClient / Google GenAI]
    CRM & Proc & Inv & Fin --> Heuristics[Deterministic Heuristics Engine]
    
    CRM & Proc & Inv & Fin --> Log[(Vanta Pilot Prediction Log)]
    CRM & Proc & Inv & Fin --> Doc[(Update Form Fields & Reload)]

---

2. Dynamic Database Configuration (Vanta Pilot Settings)

Vanta-Pilot configuration is 100% database-driven. Settings are managed via the Vanta Pilot Settings Single DocType stored in MariaDB (tabSingles), organized into a 3-tab layout:

🤖 Tab 1: AI Engine & LLM

• Gemini API Key (Password): Securely encrypted in DB.
• Default Model (Select): Options include gemini-3.5-flash, gemini-3.6-flash, gemini-flash-latest, gemini-3.1-pro-preview.
• Sampling Temperature (Float): Default 0.2 (lower temperature guarantees deterministic business analytics).
• Auto-Analysis Triggers:
  • Enable Auto-Analysis on Submit (Check): Autonomous evaluation upon doc submission.
  • Enable Nightly Self-Tuning & Learning (Check): Background model parameter recalibration.

📦 Tab 2: Procurement & Sourcing

• RFQ Routing Threshold (KES) (Currency): Default KES 50,000. PR values exceeding this trigger multi-vendor RFQ routing.
• Supplier Scoring Weights (Percent):
  • Price Competitiveness Weight (%): Default 50.0%
  • Lead Time Speed Weight (%): Default 30.0%
  • Historical Reliability Weight (%): Default 20.0%

📊 Tab 3: Sales & Finance Viability

• Minimum Healthy Margin (%) (Percent): Default 25.0%. Quotations with gross margins below this trigger warning flags and executive review requirements.

---

3. Telemetry & Auditability

Every request evaluated by Vanta-Pilot generates a record in the Vanta Pilot Prediction Log:

• Agent Type: Identifies the calling agent (e.g., CRM & Sales Co-Pilot, Procurement Co-Pilot).
• Reference DocType & Name: Clickable link to the affected record.
• AI Score & Recommendation: The computed numerical score ($0-100%$) and status (Recommended, Needs Review, High Risk).
• Raw Prompt & Raw Response: JSON logs capturing exact inputs and outputs for compliance and self-tuning.