A Singapore Government Agency Website
How to identify
A chatbot that actually works! Aunty will learn your context and guide you on the next steps to publishing an API on APEX
Chatty Aunty is the chatbot that actually works. It is built to help users with non-trivial tasks. e.g. To publish a cross-zone API on APEX. Instead of dumping documentation (or worse, links to documentation), Aunty will ask you contextual information of your current state and guides you through the next steps.
Chatty Aunty is the chatbot that actually works. It is built to help users with non-trivial tasks. e.g. To publish & consume a cross-zone API on APEX. Instead of dumping documentation (or worse, links to documentation), Aunty will ask you contextual information of your current state and guides you through the next steps.
Users who are unfamiliar with APEX & APIs take an average of 5 weeks to publish & consume a cross zone API. Most service desk interaction cycles are wasted on simple clarifications and asking for information from users. While a troubleshooting call with an APEX engineer is effective, it is manpower intensive.
Note: Measured from create date to publish date
Aunty will be able to guide user to publish a cross-zone API within 2 hours. Using agentic flows, Aunty will also help to perform simple tasks like whitelisting of IP or calling of APIs. Beyond APEX, the team feels this could pave the way of having an AI assistant for other Govtech applications like CStack or Stackops. It can also be used to be an IM assistant. Imagine a day you can ask Aunty "If I review NRIC numbers in my search portal, am I violating IM?"
The team will implement a chatbot (gpt-4o mini) with RAGflow. The key difference is that we will be using service desk tickets in addition to the documentation as inputs to the bot. We will also develop a set of best practices on how documentation should be written such that a bot can effectively learn from it. Yes thats right! Documentation is now meant for the bot to read. Very few users will read the entire documentation to find a specific solution to their own problem. They are likely to send a service desk ticket.