Most websites are static and unable to respond to user questions in real time.Businesses rely on manual support, leading to slow responses and missed opportunities.
Ernesto is an AI chatbot that allows businesses to answer user questions automatically, using their own website content.
The goal is to provide instant, accurate responses without manual intervention.
Approach
Built a functional prototype integrating LLM models with a retrieval-based system.
The chatbot uses a RAG (Retrieval-Augmented Generation) approach, combining vector search with contextual prompts to generate accurate responses based on website content.
Content can be trained directly through text inputs or structured Q&A, allowing businesses to refine how the assistant responds.
The system also supports automated actions triggered by user intent — for example, subscribing a user to a newsletter by calling external endpoints when relevant context is detected.
This approach allows the assistant to move beyond static responses, acting as an interface between users and business operations.
Admin Dashboard
To make the assistant usable in real-world scenarios, I designed a management dashboard where teams can train, monitor, and improve responses.
I combined design and code throughout the process, using tools like Figma Make and React to rapidly prototype and validate ideas. This allowed me to test real interactions early and iterate based on behavior, not just visuals.
Next Steps
Expanding the assistant’s capabilities with deeper integrations and real-time data access, enabling it to move beyond static responses and act as an interface between users and business operations.
Further refining the admin experience to help teams better understand performance, identify gaps, and continuously improve responses.
