A developer perspective on building an AI-powered prevention science assistant
It started as a fun prototype. The goal was simple: could I build a quick dashboard and chatbot with added “prevention context” that integrates with different providers? What began as a holiday experiment quickly evolved into something more substantial. On a prevention side, Matej Košir from Institute Utrip, gave a professional guidance for the “feel” and correct prevention context for the chatbot, that is based on science based prevention research. With that, chatbot, even at the current, early state, is useful for giving specific or general answer regarding prevention.
RAG and the Quest for Better Context
A key focus was ChromaDB with RAG (Retrieval Augmented Generation). I experimented with different settings to find the best “chunks” of context—all with the aim of testing whether prepared content could be successfully integrated into the chatbot. The challenge wasn’t just connecting the pieces; it was making the retrieval meaningful enough that the bot could answer real questions.
From Prototype to Working Product
I think the prototype more or less succeeded. The result is a chatbot that can answer questions on prevention science, policy, and practice. PreventAdvisor is a niche tool, and hopefully the community will find it useful and provide feedback on missing features.
The current implementation includes a simple feedback system, session saving, and everything packaged in a custom WordPress theme—enough to test whether the idea is market-ready. In the next major version, this could move to a more serious tech stack, but first we’ll see if the implementation is “good enough” to be used by the community. The chat service has big potential as it can be augmented with different tools, functions etc.
The chatbot “PreventAdvisor” was recently launched as a V1.0 and can be tested for a few €.