RadarBot · Radar Estate

Production-style portfolio project: an AI real estate chatbot for the fictional brokerage Radar Estate.

RadarBot answers natural-language questions by retrieving grounded facts from a property database. It does not invent listings, prices, or availability.

Features

• FastAPI backend with SQLAlchemy + Pydantic
• SQLite by default (PostgreSQL-ready)
• Hybrid search: structured filters + Sentence Transformers + FAISS
• Conversational memory (follow-ups like “Which one is the cheapest?”)
• Ollama-first LLM integration (OpenAI-compatible fallback)
• Admin dashboard (add / edit / delete / upload images)
• Favorites, compare, mortgage calculator, schedule viewing, recommendations
• Modern responsive UI with dark mode, chat bubbles, typing indicator, voice I/O hooks
• Docker + Pytest

Project structure

radarbot/
├── backend/          # FastAPI app, config, dependencies
├── chatbot/          # Intent parsing, memory, LLM, safety
├── database/         # Session, schema.sql, seed, sample data
├── embeddings/       # Encoder + FAISS index
├── frontend/         # HTML/CSS/JS UI
├── models/           # SQLAlchemy ORM
├── routers/          # API routes
├── schemas/          # Pydantic contracts
├── services/         # Business logic
├── scripts/          # init_db, rebuild_index
├── tests/            # Pytest suite
├── Dockerfile
├── docker-compose.yml
└── requirements.txt

Quick start (local)

1. Prerequisites

• Python 3.11+ (recommended; python3.11 -m venv .venv)
• Optional: Ollama with a chat model (ollama pull llama3.2)
• Optional heavy deps for semantic search: faiss-cpu, sentence-transformers (installed via requirements.txt)

2. Setup

cd Radarbot
python3.11 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -U pip
pip install -r requirements.txt
cp .env.example .env

Tip: If FAISS / Sentence Transformers install slowly, you can still run the app — structured search and deterministic chat replies work without them.

3. Initialize database

python scripts/init_db.py

This creates tables, seeds sample listings/agents, creates the admin user, and builds the FAISS index.

4. Run the API + UI

uvicorn backend.main:app --reload --host 0.0.0.0 --port 3000

Open:

• Home: http://127.0.0.1:3000/
• Chat: http://127.0.0.1:3000/chat
• Admin: http://127.0.0.1:3000/admin
• API docs: http://127.0.0.1:3000/docs

Default admin

• Email: admin@radarestate.com
• Password: Admin123!

LLM configuration

Local (Ollama):

LLM_PROVIDER=ollama
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=llama3.2:latest

Cloud / Render (Groq — recommended):

LLM_PROVIDER=groq
GROQ_API_KEY=gsk_...
GROQ_MODEL=llama-3.3-70b-versatile

Get a free key at console.groq.com/keys.

OpenAI:

LLM_PROVIDER=openai
OPENAI_API_KEY=sk-...
OPENAI_MODEL=gpt-4o-mini

If the LLM is offline, RadarBot still answers using deterministic, database-grounded formatters.

Deploy on Render (live URL)

Repo: github.com/Oloyededaniel/radarbot

1. Create a free Groq API key: console.groq.com/keys
2. Open Render Dashboard → New → Blueprint
3. Connect GitHub and select Oloyededaniel/radarbot
4. Render reads render.yaml and creates the radarbot web service
5. Set these env vars when prompted:
   • GROQ_API_KEY = your Groq key
   • ADMIN_PASSWORD = a password you choose (e.g. Admin123!)
6. Deploy

After deploy:

• App: https://radarbot.onrender.com (or the URL Render shows)
• Chat: /chat
• Admin: /admin (email admin@radarestate.com)

Notes:

• Free Render services sleep after inactivity — first request may take ~30–60s
• SQLite is used for the demo; data resets if the instance is fully replaced
• Cloud deploys use requirements-render.txt (no Torch/FAISS) for faster builds; chat uses Groq

API endpoints

Method	Path	Description
GET	/properties	List properties
GET	/property/{id}	Property detail
GET	/search	Filtered / hybrid search
POST	/chat	Chat with RadarBot
POST	/login	JWT login
POST	/register	Create customer account
POST	/admin/add	Add listing (admin)
PUT	/admin/update?property_id=	Update listing (admin)
DELETE	/admin/delete?property_id=	Delete listing (admin)
POST	/admin/upload	Upload image (admin)
POST	/compare	Compare listings
POST	/mortgage	Mortgage estimate
POST	/viewing	Schedule viewing
POST	/email-agent	Queue agent email
GET/POST/DELETE	/favorites	Favorites
GET	/recommendations/{id}	Similar homes

Example chat prompts

• What houses do you have in Vancouver?
• Show me all apartments under $700,000.
• Which properties have 4 bedrooms?
• Do you have homes with swimming pools?
• What is Listing 1045?
• Compare Listing 101 and Listing 215.
• I need a family home.
• I'm a student.
• Which one is the cheapest? (follow-up)

Docker

docker compose up --build

The API serves on port 3000. Ollama on the host is reachable via host.docker.internal.

PostgreSQL optional profile:

docker compose --profile postgres up --build

Then set:

DATABASE_URL=postgresql+psycopg2://radarbot:radarbot@db:5432/radarbot

Tests

pytest -q

Safety principles

1. Retrieve from the database first.
2. Format or generate answers only from retrieved rows.
3. If missing, reply:
   I couldn't find that information in the Radar Estate listings.

Deployment notes

1. Prefer Groq on Render (LLM_PROVIDER=groq).
2. Set a strong SECRET_KEY (Render can auto-generate).
3. Use PostgreSQL for durable production data if needed.
4. Keep admin credentials and API keys in Render env vars only — never commit .env.
5. Optional local semantic search: pip install -r requirements.txt (includes FAISS + Sentence Transformers).

License

MIT — portfolio / educational use.

radarbot