Tablekeeper — AI Dining & Reservation Intelligence

Tablekeeper is an AI-powered dining and reservation platform built around one principle: AI recommends. Reservation Core decides. Finding a restaurant is conversational, but booking one is transactional. AI can understand whether a guest wants a quiet anniversary dinner, a group-friendly venue, or a specific dining experience—but it should never hallucinate availability or invent a confirmed reservation. Guests describe what they want to an AI Concierge, which interprets intent, searches a grounded 24-venue catalog, applies hard constraints, and ranks suitable options. When the guest books, the authoritative Reservation Core takes over for availability, confirmation, cancellation, reservation identity, and accepted terms. PostgreSQL 17 provides the transactional boundary through database constraints, row-level security, idempotency, and exclusion-backed allocations. In our concurrency proof, 50 simultaneous confirmations produced exactly 1 winner, 49 safe conflicts, 0 double bookings, and 0 server errors. Tablekeeper also provides Dining Pass, Dining Placement, venue-scoped staff operations, and a reproducible Docker judge environment. The system was developed through a BAND-based agentic software factory with specialized agents for architecture, contracts, reservation engineering, retrieval, resilience, experience, and independent verification. Work progressed through explicit handoffs and evidence gates rather than a single coding-agent loop. We also adversarially tested the finished system. Five targeted mutations attempted to break allocation lifecycle, idempotency, row-level security, trusted authentication context, and migration integrity. All five were detected, while the untouched production baseline passed 62/62 tests, followed by independent verifier and Evidence Auditor review. Tablekeeper gives guests the convenience of AI without asking them to trust AI with the truth.