In progress
Customer service chatbot (thesis)
A web chatbot for JakraCampGear customer service. The LLM uses function calling to read stock and prices from the real inventory data instead of making them up. Built on top of the Jakra system with a five-layer architecture, following the Prototyping method.
LLM function calling, Go, PostgreSQL, React
The problem
Customers often ask about stock through WhatsApp, and an admin has to check and reply by hand. Replies slow down as orders grow, and a plain LLM chatbot has no access to real stock data.
My role
This is my undergraduate thesis. I handle the research, design, and implementation.
How it works
The LLM uses function calling to decide when to query the inventory, so stock and prices come from real data instead of being made up.
The design uses a repository, service, tool, and handler pattern on top of the Jakra system, developed with the Prototyping method.
Key features
- Tools for rental stock, sale stock, prices, and product descriptions
- Web chat widget for customers
- Planned testing of function call accuracy and answer accuracy against real data
Still in progress. The scope covers questions about stock, prices, and product descriptions, in Indonesian, without payment or booking.