<aside> 💡 Version 3.4.0 brings administrator tools for Gemini, rebuilds knowledge base search and function calling, and adds a bot module generator. This version also introduces automated conversation analysis and instant feedback from students.
</aside>
Version 3.4.0 adds administrator tools for Gemini (previously only available for OpenAI) and rebuilds several core AI capabilities introduced with Gemini in version 3.3.0.
The knowledge base search (RAG) has been rebuilt with our own implementation on Google Enterprise Agent Platform. The bot now decides when to search knowledge bases on its own, can query multiple documents at once, and forms answers from across your materials. Function calling has also been rebuilt: the bot can now chain multiple actions in a single turn. For example, it can search a knowledge base and file a support request in one go, without waiting for additional user input. The overall prompt structure has been overhauled as well, resulting in cleaner responses with less unnecessary formatting and better adherence to module-specific instructions.
A new bot module generator allows coordinators to create a working bot module by describing the use case, audience, and optional conversation flow. Instead of writing instructions manually, the generator produces an initial draft with module instructions, support topics, and connected knowledge bases to start from and refine. Coordinators can also now set the bot's conversational tone (Professional or Casual), and Annie is now aware of the current date, supporting use cases that depend on dates such as deadlines or upcoming events.
Support bot conversations are now analyzed by a separate analysis agent that evaluates how the bot performed, extracts discussion topics, and determines outcomes. In addition, students who interact with Annie are now asked directly whether the conversation was helpful. Together, the automated analysis and student feedback provide insights to which modules work well, what support students find helpful, and where to improve.
Rebuilt knowledge base search with multi-document support AD-1006
The knowledge base search has been rebuilt using our own RAG implementation on Google Enterprise Agent Platform, replacing the previous approach. The bot now decides on its own when to search and can query multiple documents at once to form more complete answers.
Rebuilt function calling with action chaining AD-1007
Function calling has been rebuilt to support chaining multiple actions in a single turn. For example, the bot can search a knowledge base and file a support request in one response without waiting for additional user input.
Bot module generator AD-979
A new bot module generator allows coordinators to create a working bot module by describing the use case, audience, and optional conversation flow. Instead of writing instructions manually, the generator produces an initial draft with module instructions, matching support topics, and connected knowledge bases to start from and refine.
Customizable conversational tone AD-1022
Coordinators can now set the bot's communication style to Professional or Casual from the admin console.
Instant feedback from students AD-1003
After a conversation ends, students who interacted with Annie are now asked directly whether the conversation was helpful. This feedback can be used alongside automated analysis to identify which modules work well and where to improve.
Session-based conversation grouping AD-999
Conversations are now grouped into discrete sessions based on inactivity thresholds, forming the basis for automated analysis and instant feedback.
Automated session performance analytics AD-1012
A dedicated AI agent now automatically analyzes completed sessions to evaluate conversation success, extract discussion topics, and review bot performance.
Time-aware AI responses AD-1016
Annie now knows the current date, enabling use cases where date awareness matters, such as referencing deadlines, upcoming events, or time-sensitive instructions.
Localized support topic taxonomy AD-1018
Support topics and codes have been updated to a localized taxonomy, improving language support across the student app and administrator console.
Enhanced conversational quality and instruction adherence AD-978
The bot responds more naturally, formats answers cleanly without excess markdown, and strictly follows module-specific support rules.
Gemini support in the administrator console AD-1001
Coordinators can now configure Gemini-powered bots, upload knowledge base materials, and test modules directly in the administrator console.
New API endpoints for managing AI agents AD-1005
Developers can now programmatically create, read, update, and delete AI agents via the API.
New paginated API for retrieving session data AD-1020
Developers can now efficiently fetch session analytics and history using cursor-based pagination and filtering.
Student picker table no longer slows down the interface AD-851
Introduced async loading of student data in the background to fix severe performance drops when using the survey editor in the admin console.