Project description
We run an online store with a small support team, and our operators are drowning in repetitive questions about order status, shipping times, returns and product specs. We want to take the routine off their plate by adding an AI-powered chatbot that can answer customers directly from our knowledge base and FAQ, without making people wait in a queue for something a document already explains.
The bot should run on a modern LLM and pull answers from our existing help articles and product data, so it stays accurate instead of inventing facts. Just as important is knowing when to step back: if a conversation gets emotional, involves a refund dispute, or the model isn't confident, it should hand the conversation over to a live operator, with the full chat history attached, so the customer never repeats themselves. It should live inside the widget already on our site, matching its look rather than appearing as a separate pop-up.
We're open to discussing the model and architecture with the right freelancer, but response speed, cost per conversation, and a way to review wrong answers matter a lot to us.
What we expect from the contractor:
— set up an LLM chatbot with retrieval from our knowledge base (RAG or similar);
— configure clear escalation rules with full context handoff;
— integrate the bot into our current widget without breaking chat flows;
— provide basic analytics: unanswered questions, escalation rate, response time;
— deliver a short guide so our team can update the knowledge base themselves.
The bot should run on a modern LLM and pull answers from our existing help articles and product data, so it stays accurate instead of inventing facts. Just as important is knowing when to step back: if a conversation gets emotional, involves a refund dispute, or the model isn't confident, it should hand the conversation over to a live operator, with the full chat history attached, so the customer never repeats themselves. It should live inside the widget already on our site, matching its look rather than appearing as a separate pop-up.
We're open to discussing the model and architecture with the right freelancer, but response speed, cost per conversation, and a way to review wrong answers matter a lot to us.
What we expect from the contractor:
— set up an LLM chatbot with retrieval from our knowledge base (RAG or similar);
— configure clear escalation rules with full context handoff;
— integrate the bot into our current widget without breaking chat flows;
— provide basic analytics: unanswered questions, escalation rate, response time;
— deliver a short guide so our team can update the knowledge base themselves.
Web development Cash, Bank transfer
