Founded and independently built Solandra, an AI-powered customer support platform that answers visitor questions from a business's documentation and website content, with source-backed responses. I took it from architecture to a working private beta, and it is now live with paid plans.
Setting up an AI support bot usually means writing and maintaining a knowledge base by hand, and the questions the bot can't answer vanish without anyone on the team seeing them.
Solandra crawls the business's website, ingests the content into a multi-tenant store, and answers through hybrid retrieval (Qdrant vectors + PostgreSQL full-text) with LLM answer generation that cites its sources. Questions it couldn't answer feed a knowledge-gap list so the team knows what to document next.
Hybrid retrieval: Qdrant vector search + PostgreSQL full-text
WhyAnswers have to come from the business's own content and point back to it, so retrieval quality decides answer quality.
ApproachCombines Qdrant vector search with PostgreSQL full-text search, then generates the answer with an LLM over the retrieved content and returns the source pages alongside it.
Unanswered questions become a review list
WhyWhen a support bot can't answer a question, the question usually just disappears, and the gap in the documentation never gets fixed.
ApproachKnowledge-gap detection flags the questions Solandra couldn't answer and surfaces them to the business as a list to improve its documentation.
- →Hybrid retrieval combining Qdrant vector search, PostgreSQL full-text search, and LLM answer generation
- →Embeddable support chatbot with conversational context, configurable widget settings, and human escalation
- →Automated knowledge-gap detection that turns unanswered questions into a documentation review list
- →Website crawling, content ingestion, and multi-tenant data management
- →Secure API integrations for the embeddable widget