Back to projects Project · AI Agent

AI Appointment Scheduler

A full-stack, AI-powered scheduling system for a hair salon: customers book appointments through a natural-language chat interface instead of a traditional form.

Agentic AI LangGraph ReAct Chat booking

Source code Get in touch

What it does

Customers describe what they need in plain English (“I'd like highlights next Saturday at 2pm”) and a conversational AI agent handles the entire booking flow — checking availability, confirming details, and optionally collecting payment — all in a single chat thread.

Key engineering highlights

LangGraph ReAct agent

A tool-calling loop (check availability → book → request payment), compiled once at startup and shared across all concurrent sessions via checkpoint-keyed conversation state.

Dual request-path architecture

The agent handles customer interactions while admin reads bypass the LLM entirely for direct, low-latency service access.

Structured payment trigger

The agent emits a structured sentinel via a tool response; the backend extracts it and the frontend renders an inline payment modal — with no client-side prompt parsing.

Clean service boundary

All persistence logic lives in a single service class, making the storage layer swappable without touching the agent or the API.

Natural-language dates

With python-dateparser, the agent passes the user’s phrasing verbatim to tools rather than forcing ISO format on the customer.

Architecture

┌──────────────────────────────────────────────────────────────┐
│                          BROWSER                             │
│   ┌────────────────────────┐   ┌────────────────────────┐    │
│   │   client-app  :4200    │   │   admin-app  :4201     │    │
│   │   SchedulingPrompt     │   │   AppointmentList      │    │
│   │   • chat UI            │   │   • live table         │    │
│   │   • threadId (UUID)    │   │   • loading / error    │    │
│   │   • payment modal      │   │                        │    │
│   └───────────┬────────────┘   └───────────┬────────────┘    │
└───────────────│────────────────────────────│────────────────┘
        POST /chat                    GET /appointments
        POST /appointments/{id}/pay   POST /appointments/{id}/pay
                │                            │
┌───────────────▼────────────────────────────▼────────────────┐
│                       FastAPI  :8000                         │
│   ┌────────────────────────┐   ┌────────────────────────┐    │
│   │        chat.py         │   │     appointments.py    │    │
│   │   rate limiter         │   │   GET  /appointments   │    │
│   │   (20 req/60s/thread)  │   │   GET  /…/{id}         │    │
│   │        │               │   │   POST /appointments   │    │
│   │        ▼               │   │   POST /…/{id}/pay     │    │
│   │   graph.ainvoke()      │   └───────────┬────────────┘    │
│   │   scan ToolMessages    │               │                 │
│   │   → extract            │               │                 │
│   │     payment_request    │               │                 │
│   └───────────┬────────────┘               │                 │
└───────────────│────────────────────────────│────────────────┘
                ▼                            │
┌───────────────────────────────┐            │
│      LangGraph StateGraph     │            │
│    (compiled once at startup) │            │
│    ┌─────────┐  tool_calls?   │            │
│    │  agent  │───────────────►│            │
│    │  node   │◄───────────────│            │
│    └────┬────┘   ┌─────────┐  │            │
│         │        │  tools  │  │            │
│         ▼        │  node   │  │            │
│        END       └────┬────┘  │            │
│    MemorySaver        │       │            │
│    (thread_id key)    │       │            │
└───────────────────────│───────┘            │
                        ▼                    ▼
┌──────────────────────────────────────────────────────────────┐
│                     AppointmentService                        │
│                (single owner of all storage)                  │
│   check_availability │ book_appointment │ cancel │ mark_paid  │
│                _store: dict[id → Appointment]                 │
│                      (in-memory, PoC)                         │
└───────────────────────────────┬──────────────────────────────┘
                                │  (DATABASE_URL wired, not yet used)
                                ▼
                       ┌──────────────────┐
                       │   PostgreSQL     │
                       │   port 5433      │
                       │   (Docker)       │
                       └──────────────────┘

The key architectural insight is the payment trigger: the agent emits a payment_request sentinel via a tool response, extracted on the backend — the frontend renders a payment modal without ever parsing the LLM’s text.

Tech stack

AI / Agent

LangGraph (ReAct)
LangChain
Claude — Anthropic
Custom agent tools

Backend

Python
FastAPI
Pydantic Settings
python-dateparser

Frontend

Angular 21
TypeScript
Signals reactivity
ngx-markdown

Infrastructure

Docker
PostgreSQL

Want an AI agent for your business?

Let's talk about how I can design conversational AI agents that automate your customer journeys.

Get in touch Source code