I’m excited to introduce the latest version of LingoStand, packed with enhancements to the Scenario Creation Agent for faster and more efficient scenario generation. My goal is to streamline the experience, making it easier and quicker for users to dive into relevant simulations. At the beginning of this update, I conducted usability tests with some users, whose feedback was invaluable in refining these improvements. Here’s what’s new:
🚀 Enhanced Scenario Creation
More Fluid Scenario Input
- Loosely Written Scenarios: Now, you can provide just an idea, and the agent will skip guidance questions if all necessary elements are detected, directly generating a scenario.
- Shorter Initial Messages: The agent’s first message is now more concise and to the point, with examples included upfront. This change addresses feedback that longer introductions were often skimmed.
- Guided Specific Interests: If you enter a broad interest (e.g., “music”), the agent now helps refine it to something more specific for better scenario generation.
- Pasting Entire Scenarios: You can copy and paste a previously generated scenario into the Scenario Creator to retry or share it.
- Interest Combinations for Creativity: You can combine interests to generate unique scenarios. For example:
- “Cooking” + “Star Trek” → A chef aboard a spaceship.
- “Negotiation” + “Diablo video game” → A merchant in Tristram.
- Personalization Data Usage: If you provide minimal input, the agent will pull from profile data (e.g., interests, focus areas) to suggest scenario ideas.
📝 Improved User Experience
More Actionable, Concise Feedback
- Both the Scenario Creation Agent and Evaluator Agent now provide shorter, more actionable responses, focusing on key takeaways instead of lengthy explanations.
- Removed unsupported end-of-scenario options, including “Restart scenario” and “Continue scenario.” Now, you can only:
- Ask for more sentences to continue practicing.
- Request performance feedback.
🐞 Bug Fixes & Internal Improvements
- Game Master Agent: Fixed cultural corrections that kept the flow from moving forward
- Resolved Game Master role confusions in some scenarios
- Improved HTML formatting issues in outputs.
- Internal updates: I introduced a new process to compare LLMs for specific agents, allowing me to find the best-performing model and fine-tune it for better accuracy and consistency.
This update makes scenario creation smoother, more intuitive, and personalized, helping you focus on language practice rather than setup. A big thank you to all the users who participated in the usability tests—your feedback was incredibly helpful! Let me know what you think, and happy learning! 🎉