Inspiration
The inspiration for the Gemini Plugin came from the limitations we observed in traditional chatbots, which often struggle to strike a balance between dynamic conversation and maintaining control over specific user flows. We envisioned a solution that could blend the reliability of structured interactions with the adaptability of AI, empowering chatbots to offer a truly engaging and human-like experience. Our goal was to bridge this gap by introducing a "dual intelligence" model—one that allows chatbots to leverage the best of both worlds.
What it does
The Gemini Plugin enhances Hexabot's capabilities by enabling a unique dual intelligence approach. It provides the chatbot with two modes of thinking: structured, rule-based logic for precise workflows, and generative AI for more open-ended, flexible conversations. This combination allows the chatbot to maintain control during crucial processes, like customer support or data collection, while still being capable of engaging dynamically to handle unexpected user queries or create personalized experiences.
How we built it
We built the Gemini Plugin on top of the Hexabot architecture, which leverages NestJS for the API and React for the frontend. For the AI component, we integrated a BERT-based model for natural language understanding alongside OpenAI's GPT model for generative responses. This setup allows Gemini to identify when a conversation needs to switch between structured and generative modes, using contextual cues and user input as triggers.
Our development process was highly iterative, involving extensive testing to ensure a seamless transition between the rule-based and generative responses. We employed Docker Compose for environment consistency, making it easy to test interactions in different configurations.
Challenges we ran into
One of the biggest challenges we faced was achieving balance between structured logic and generative AI. We wanted to ensure that the chatbot remained consistent and reliable in critical contexts, while still being capable of dynamic interactions. Finding the right heuristics for switching modes was a complex task that required a lot of experimentation.
Another challenge was managing latency and response times, especially when calling external AI APIs. We worked to optimize the architecture so that the end-user experience remained smooth and responsive even with AI-in-the-loop.
Accomplishments that we're proud of
We are proud of how the Gemini Plugin seamlessly integrates two distinct conversation approaches into a single, cohesive experience. It was incredibly rewarding to see our chatbot navigate a controlled workflow and then effortlessly pivot to a more creative, open-ended conversation when prompted by the user.
We’re also proud of the community engagement our project has attracted—developers are already contributing ideas for improving the plugin, and the open-source nature of Hexabot has sparked a lot of interest.
What we learned
We learned a lot about the intricacies of conversational AI, particularly how to manage expectations around generative AI outputs and balance them with predictable, rule-based responses. We also gained deeper insights into the importance of managing context effectively to ensure that conversations remain coherent, regardless of how often they switch between different interaction modes.
Another key learning was the importance of community involvement. Open-source projects thrive on collaboration, and engaging early with users and developers led to invaluable feedback that shaped our final product.
What's next for Dynamic AI-Powered Chatbots with Hexabot Gemini Plugin
The next step for the Gemini Plugin is to refine the intelligence switching heuristics further and introduce more advanced context management capabilities. We plan to enhance integration with more messaging platforms, making it easier to deploy across different channels seamlessly. Additionally, we want to introduce a self-learning mechanism so that the chatbot continuously improves its context-switching capabilities based on user feedback.
We’re also looking to further engage the open-source community to add new features, contribute to the plugin’s growth, and ensure it evolves to meet real-world needs. Hackathons like this are just the beginning, and we’re excited to see how Gemini will inspire new kinds of AI-powered conversational experiences.
Log in or sign up for Devpost to join the conversation.