Back to projects

Solis

Development of the Solis interface, an AI agent in Landingi that analyzes EventTracker data and landing page context, automatically generating insights, alerts and optimization recommendations. Users can also ask questions about their page results through an AI chat.

Solis chat with a landing page stats summary in a table – enlarge

Where it started

EventTracker gave users detailed data about visitor behaviour, but interpreting it required manually analysing charts and events. Solis was meant to simplify that process by delivering automatic insights and recommendations and letting users ask questions about their landing page results.

Ready-made insights were already available in the app, but there was no interface for having a conversation with the AI agent.

The challenge

Together with another frontend developer I was responsible for building the Solis chat in the SPA. My main focus was API integration, handling streamed responses, managing conversation state and history, and keeping the interface smooth. I was also responsible for migrating and developing the existing insights views.

How I built it

I built the chat interface using useChat from the Vercel AI SDK, integrating response streaming from the Mastra API and JWT authentication. I implemented generation state handling, progressive response rendering, request cancellation and retry, and Markdown formatting including tables.

I managed scroll behaviour during streaming so users could follow the response being generated while still being able to scroll back to earlier messages without being forced to the bottom. I added conversation restoration when the panel is reopened or the page is refreshed — the frontend fetched the conversation ID and its history from the API, replayed the messages and allowed continuing the same thread.

I also developed new conversation handling, message suggestions, response rating and copying, error handling, access control and Mixpanel analytics events.

I moved the existing insights to the Solis panel, adding type filtering, an unread counter and rating. I then built a separate Insights page with a detail view, insight deletion with a reason, and navigation to the related landing page. I covered key functionality with automated tests in Vitest and Testing Library.

The outcome

The result is an integrated Solis interface where users can browse automatically generated insights and recommendations and analyse their landing page results through a conversation with the AI agent. Streaming lets users follow responses as they are generated, and conversation history restoration means they can continue a previous analysis without starting over.

The expanded insights views bring recommendations, their details and related landing pages together in one place.

Any questions?

Get in touch