Aleysia: NLP property chatbot

Undergraduate thesis. A chatbot for a housing estate that takes a request in plain Indonesian, recommends units from it, handles bookings, and simulates mortgage (KPR) offers.

The problem

Property searches assume the visitor already knows the vocabulary: bedrooms, building type, payment scheme. People who genuinely need to find a home often cannot describe what they want in the form's terms.

What I did

• Built the intent classifier on TF-IDF features with logistic regression, trained against an Indonesian housing-domain corpus.
• Added regex entity extraction so the slots (budget, location, bedroom count) come back typed instead of as free text.
• Wired Sastrawi stemming and NLTK for Indonesian language preprocessing.
• Implemented unit recommendation, session-based booking, and KPR simulation across partner banks.
• Stored state in Supabase/PostgreSQL and exposed the API with PyJWT auth, packaged for deploy with Docker Compose.

Impact

78

commits on the backend

Intents

classified from TF-IDF + logistic regression

KPR

mortgage simulation included

Links

Stack

• Python
• Flask
• scikit-learn
• NLTK
• Sastrawi
• Supabase
• PostgreSQL
• Gemini
• Docker
• pytest