Casa Alo's Bistro — Aria AI Chatbot

Demo 1 of the AI Alo Portfolio · aialo.io

A RAG-powered restaurant chatbot built with FastAPI, ChromaDB, and sentence-transformers. Supports Ollama (local, free) and Claude API (cloud, best quality) via a single .env flag.

The pain point it solves: Restaurant staff spend hours answering the same questions every day — menu questions, hours, allergens, reservations. Aria handles all of it 24/7.

---

System Architecture

flowchart TD
    subgraph Browser["Browser (Client)"]
        SITE["Restaurant Website\nindex.html + app.js\nAria chat widget"]
    end

    subgraph Server["FastAPI Server — port 8080"]
        MAIN["main.py\nStatic file serving + CORS"]
        ROUTER["router.py\nPOST /api/chat"]
    end

    subgraph RAG["RAG Pipeline"]
        direction TB
        EMBED["embedder.py\nsentence-transformers\nall-MiniLM-L6-v2 (local)"]
        CHROMA[("ChromaDB\nchroma_store/\npersistent vector DB")]
        PIPE["pipeline.py\nSystem prompt + context\n→ Aria personality"]
    end

    subgraph LLM["LLM Layer (dual)"]
        CLAUDE["☁ Claude Haiku\n(Anthropic API)"]
        OLLAMA["⬡ Ollama llama3.2\n(local fallback)"]
    end

    subgraph Data["Knowledge Base"]
        FILES["data/\nmenu.json · faqs.json\nrestaurant_info.json"]
    end

    INGEST["scripts/ingest_data.py\n(run once)"]

    SITE -->|"POST /api/chat"| ROUTER
    MAIN --> ROUTER
    ROUTER --> PIPE
    PIPE --> EMBED
    EMBED -->|"query embedding"| CHROMA
    CHROMA -->|"top chunks"| PIPE
    PIPE -->|"API key set?"| CLAUDE
    PIPE -->|"no API key"| OLLAMA
    CLAUDE -->|"response"| ROUTER
    OLLAMA -->|"response"| ROUTER
    FILES --> INGEST
    INGEST --> EMBED
    EMBED -->|"batch embeddings"| CHROMA

Loading

---

Part of the AI Alo Portfolio

Demo	Business	Key Feature
Demo 1	Casa Alo's Bistro	Restaurant RAG chatbot
Demo 2	Rappahannock Realty Group	Lead qualifier + CRM dashboard
Demo 3	Luminara Med Spa	Treatment recommender + candidacy screening

Built by Aloysious Kabonge — AI automation consulting for local businesses in Fredericksburg, VA.

---

Stack

Layer	Tech
Backend	FastAPI + Uvicorn
Vector DB	ChromaDB (local, persistent)
Embeddings	sentence-transformers all-MiniLM-L6-v2
LLM (cloud)	Anthropic Claude (haiku)
LLM (local)	Ollama (llama3.2)
Frontend	Vanilla HTML/CSS/JS (embeddable widget)

Quick Start

1. Install dependencies

cd casa_alos_bistro
pip install -r requirements.txt

2. Configure environment

cp .env.example .env
# Edit .env — add ANTHROPIC_API_KEY if you have one, or leave blank to use Ollama

3. Ingest data into ChromaDB

python scripts/ingest_data.py

This runs once (or whenever you update the data files). Downloads the embedding model on first run (~90MB).

4. Start the server

uvicorn app.main:app --reload

Open http://localhost:8000 — you'll see the restaurant website with the chat widget.

API Endpoints

Method	Path	Description
POST	/api/chat	Send a message, get a response
GET	/api/health	Check status, provider, doc count
DELETE	/api/session/{id}	Clear a conversation session

Chat request

POST /api/chat
{
  "message": "What are your most popular dishes?",
  "session_id": "user_abc"
}

Chat response

{
  "response": "Our guests always rave about the Cacio e Pepe...",
  "sources": ["menu/Paste (Pastas)", "faqs/Menu & Food"],
  "session_id": "user_abc",
  "provider": "claude"
}

LLM Selection Logic

ANTHROPIC_API_KEY set?
  ✓ → Uses Claude (cloud, best quality)
  ✗ → Uses Ollama (local, free, requires Ollama running)

Embeddings always run locally via sentence-transformers — no API key needed regardless.

Data Files

File	Contents
data/menu.json	Full menu — 7 categories, 40+ items, prices, dietary info
data/faqs.json	30+ FAQ Q&A pairs across 6 categories
data/restaurant_info.json	Hours, address, contact, parking, policies

To update content, edit the JSON files and re-run python scripts/ingest_data.py.

Project Structure

casa_alos_bistro/
├── app/
│   ├── main.py          # FastAPI app + CORS + static file serving
│   ├── router.py        # Chat, health, session endpoints
│   ├── config.py        # Settings (pydantic-settings, .env)
│   ├── models.py        # Request/response schemas
│   └── rag/
│       ├── embedder.py  # ChromaDB collection setup
│       ├── retriever.py # Top-k semantic search
│       └── pipeline.py  # RAG: retrieve → generate (Ollama or Claude)
├── data/
│   ├── menu.json
│   ├── faqs.json
│   └── restaurant_info.json
├── frontend/
│   └── index.html       # Restaurant website + embedded chat widget
├── scripts/
│   └── ingest_data.py   # One-time data indexing script
├── requirements.txt
├── .env.example
└── .gitignore