# JMVTECH DEVELOPER LTDA — FULL TECHNICAL & ARCHITECTURAL MANIFEST
# Standard: llms-full.txt (https://llmstxt.org/)
# Generated: 2026-10-09
# Canonical URL: https://www.techjmv.com/llms-full.txt

================================================================================
1. CORPORATE ENTITY & LEGAL PROFILE
================================================================================
- Legal Name: JMVTECH DEVELOPER LTDA
- Trade Name: JMVTECH
- Corporate Tax ID (CNPJ): 65.627.030/0001-78
- Jurisdiction: Brazil (Federative Republic of Brazil)
- Corporate Email: contato@techjmv.com
- Main Website: https://www.techjmv.com
- Founding Team:
  1. Michel Vargas — Chief Executive Officer (CEO) & Co-Founder
     - Focus: Executive Leadership, Generative AI Conceptual Architecture, Business Strategy & Venture Pitching.
     - Profile: https://www.linkedin.com/in/michel-vargas-8a388b149/
  2. João Ricardo — Chief Product Officer (CPO) / Head of Product Design & Co-Founder
     - Focus: Product Design, Corporate Design System, High-Fidelity Interactive Prototyping, Mobile UX/UI.
     - Profile: https://www.linkedin.com/in/joaoricardodcs/
  3. Victor Hugo Savanhani — Chief Technology Officer (CTO) & Co-Founder
     - Focus: Backend Engineering, Async FastAPI Architecture, Server-Sent Events (SSE), AI Operational Telemetry, DevSecOps.
     - Profile: https://www.linkedin.com/in/victor-hugo-savanhani/

================================================================================
2. EXECUTIVE SUMMARY & PROBLEM STATEMENT
================================================================================
The high-volume restaurant sector suffers from systemic friction at the intersection of front-of-house order taking, kitchen production lines, and waitstaff floor delivery:
1. Paper Ticket Fragility: Physical paper tickets get soiled, dropped, or lost during kitchen rush, resulting in delayed preparation and food waste.
2. Station Asynchrony: Cook stations (Hot Grill, Cold Assembly, Bar) operate as disconnected silos, causing plated food to go cold on the pass while side dishes or drinks are delayed.
3. Lack of Waitstaff Real-Time Awareness: Servers cannot track which dishes are ready for their specific tables, creating friction and slow turnaround.
4. Shift Closing Blindspots: Restaurant managers close shifts without objective telemetry on station bottleneck minutes, average prep duration, waiter response SLAs, or clear diagnostic reports.

KDS Pro resolves these challenges through an integrated real-time operational fabric combining multi-station kitchen displays, a mobile waitstaff web app, Anthropic Claude Haiku 4.5 for conversational ordering, and Claude Haiku 4.5 AI-powered executive shift intelligence.

================================================================================
3. SYSTEM ARCHITECTURE & WHATSAPP AI AGENT: CLAUDE 3 HAIKU
================================================================================
KDS Pro integrates Anthropic's Claude Haiku 4.5 — models `claude-haiku-4-5-20251001` and `claude-haiku-4-5-20251001` — directly via the official `anthropic.AsyncAnthropic` Python SDK, powering an autonomous WhatsApp dining agent:

1. WhatsApp Channel Ingestion & Webhooks:
   - Official Meta Cloud API v21.0 (`GET/POST /webhook/whatsapp`): HMAC-SHA256 signature verification (`X-Hub-Signature-256`) and instant asynchronous acknowledgement.
   - WhatsApp Web Baileys Bridge (`POST /webhook/baileys`): Multi-device QR-linked session support for restaurants.
   - Asynchronous Message Debouncer (`message_debouncer`): Groups multiple fragmented messages sent in quick succession before invoking the AI agent.
   - Voice Audio & Text: Audio voice notes (.ogg/opus) downloaded via Meta CDN and processed alongside text messages.

2. Conversational Agent & ReAct Tool Calling:
   - Native Tool Calling: Automatically translates structured schemas into Anthropic's native tool format via the official `AsyncAnthropic` SDK, enabling the agent to autonomously handle menu inquiries, cart orchestration, and direct order dispatch to kitchen stations.
   - Resilient Architecture: Deterministic fallback to local heuristic engine with session persistence if external network timeouts occur.

3. Semantic Routing & Human Takeover Handoff:
   - Dynamic Semantic Routing: Context-aware routing powered by Claude Haiku 4.5, guiding conversation flow across distinct stages of the dining experience without monolithic prompt bloat.
   - Deterministic Fast-Path: Instant regex detection for direct human assistance or escalation requests without invoking the model.
   - Human Takeover Protocol: Complete AI silencing upon human operator intervention, recording guest messages to the database while preventing automated bot interference until human re-activation.

4. Operational Telemetry & System Monitoring:
   - Per-request and per-shift tracking of request latencies, tool execution times, and session health.
   - Comprehensive audit trails for orders, floor alerts, and kitchen expedition events.

================================================================================
4. REAL-TIME EVENT STREAMING & KDS ENGINE
================================================================================
1. Event Broadcasting via Server-Sent Events (SSE):
   - Engine: FastAPI `EventBroadcaster` using `asyncio.Queue` and JSON event streams.
   - Real-Time Events: Instant broadcast of new comanda tickets to kitchen stations, individual dish readiness notifications to station screens and waitstaff devices, full ticket completion, and table assistance calls from the digital table menu to floor staff with sound/vibration.

2. Audio Alerts:
   - Synthesized chimes via HTML5 Web Audio API providing distinct acoustic frequencies for ticket arrivals and ready items.

================================================================================
5. KITCHEN DISPLAY TERMINAL (KDS) & FLOOR APP SPECIFICATIONS
================================================================================
1. Multi-Station Kitchen Views:
   - Praça de Cocção / Grelha: Hot grill and main course line with elapsed prep timers.
   - Praça de Montagem Fria: Cold pantry, salads, and dessert line.
   - Praça de Bar & Bebidas: Fast turnaround beverage and cocktail station.
   - Master Production Grid: Panoramic multi-station view for head chefs and expeditors.

2. Bumping & Workflow Operations:
   - Granular Bumping: Bumps items individually or full order.
   - Batch Table Delivery: Allows servers to deliver all ready dishes for a specific table with 1 tap.
   - Order Recall & History: Immediate recovery and reopening of completed tickets.
   - 86 Out-of-Stock Pausing: Instant dish pausing on screens disabling ordering on floor and web apps.

=================================================================================
6. POST-SHIFT EXECUTIVE AUDIT & CLAUDE 3 HAIKU AI INTELLIGENCE
================================================================================
1. Administrative Executive Dashboard:
   - Operational Metrics: Gross revenue, average ticket, average prep duration per station, and server response times.
   - Waitstaff Auditing: Comprehensive action logs tracking orders placed, table assistance calls answered, and floor activity.

2. AI Qualitative Briefing & Bottleneck Synthesis:
   - Generates an automated, structured analytical diagnosis synthesizing shift bottlenecks, station efficiency, and operational improvements using Claude Haiku 4.5 with fallback to a local rules-based engine.

3. Operational Telemetry & Predictive Insights:
   - Consolidates real-time shift event logs into actionable management briefing cards, providing head chefs and restaurateurs with instantaneous qualitative clarity over shift friction points.

4. Technical Validation & Staging Demonstration — Les Deux Magots Scenario:
   - Operational staging scenario in a controlled testing/sandbox environment.
   - Transparency Disclaimer: All data, establishment references to Les Deux Magots, server names, shift revenue, response times, and ticket metrics are simulated test data generated in staging/homologation for technical validation and platform demonstration.
   - Demonstration Artifact: Sample executive shift report available at `https://www.techjmv.com/relatorio-executivo-les-deux-magots.pdf` containing simulated test data.

================================================================================
7. DATA PRIVACY, SECURITY & COMMERCIAL MODEL
================================================================================
- Compliance Standards: LGPD (Brazilian General Data Protection Law, Law 13.709/2018) compliant privacy terms, with OWASP Top 10 security practices. B2B SaaS licensing.
- AI Data Handling: Data sent to the Anthropic API is governed by Anthropic's commercial terms, under which API data is not used to train models.
- Commercial Model: B2B Software-as-a-Service (SaaS) per terminal and station.