The idea for HaloSatoshi stemmed from frustration with the inefficiencies of online customer service (CS) systems in Indonesian banks. Often, bank websites lack comprehensive information, forcing users to contact CS even for general inquiries. This process is slow—users must take a ticket, wait in a queue, and endure delays as CS agents think through responses. It’s highly inefficient and time-consuming.
HaloSatoshi aims to solve this by providing a 24/7 online CS chatbot for Satoshi Bank, delivering fast, accurate, and automated responses to customer queries and requests.
HaloSatoshi is a 24/7 AI-powered chatbot service by Satoshi Bank, designed to assist customers with a wide range of requests. Whether you're seeking general information or need to report/request specific services, HaloSatoshi is ready to help with specific topics such as:
- Opening a new account
- Reporting a mistaken transfer
- Blocking an account due to repeated ATM PIN errors
- Handling failed transactions with deducted balances
- Recovering forgotten PIN/password
- Requesting a new token
- Blocking an account due to fraud
- Updating personal data (KTP, phone number, email, etc.)
- Reporting a lost or swallowed ATM card
- Responsive UI: Built with React, featuring a glassy, dark-purple theme with real-time typing effects.
- Session Management: Users can revisit past chat sessions via a sidebar with chat history.
- Out-of-Scope Detection: Rejects irrelevant queries with a custom message, guiding users to contact support or visit the website.
- RAG-Powered Responses: Uses Retrieval-Augmented Generation (RAG) for accurate, context-aware answers.
- TF-IDF and Cosine Similarity: Leverages natural language processing for efficient question retrieval.
- Frontend: React, CSS (glassy effects, gradients, animations)
- Backend: Node.js, Ollama API for more natural response generation
- NLP: RAG, TF-IDF, Cosine Similarity (via
naturallibrary) - State Management: React Hooks (
useState,useEffect) - API Integration: Fetch for frontend-backend communication
Due to GitHub repository upload limitations, the RAG model files are not included here.
For more information, please contact the creator/developer.