TableTalk AI

A hackathon-ready restaurant host/concierge that improves over time using evals + policy patching, and supports voice playback via OpenAI TTS.

Features

• Host Assistant: Handles wait times, hours, location, reservations, menu questions, allergies, pickup status, and basic complaints.
• Self-Improving Loop (planned): Evaluator scores output and violations, Patcher updates "Host Policy", and next replies use the latest policy.
• Voice Output: Uses OpenAI TTS for assistant replies.
• Restaurant Management: Manage restaurant profiles, menus, and view policy history.
• PWA Support: Web app manifest for installability on supported browsers (e.g. Add to Home Screen).

Tech Stack

• Framework: Next.js 14+ App Router + TypeScript
• Styling: TailwindCSS + shadcn/ui
• Database: Prisma + SQLite
• LLM: OpenAI-compatible endpoint (GPT-4o-mini default)
• Voice: OpenAI TTS
• Observability: Pino logs

Setup

1. Clone the repository.
2. Install dependencies: pnpm install.
3. Copy .env.example to .env and fill in the required variables.
4. Run database migrations: pnpm prisma migrate dev.
5. Start the development server: pnpm dev.

Environment Variables

See .env.example for the list of required environment variables.

Planned

• Self-improving loop: Evaluator to score assistant output and policy violations, plus a Patcher to update the Host Policy from low scores; new replies would then use the updated policy. The app already stores policy versions and messages; this would add the evaluation and policy-patch pipeline.