Project Description
I’m enhancing my existing SaaS platform with an AI-powered text chat agent that sits inside our website’s live-chat widget. The core objective is simple: when a user opens the chat on our site, the agent should be able to converse naturally, resolve routine queries, and seamlessly hand off anything it can’t solve to the human team.
Here’s how I picture the workflow:
• A visitor types a question into the live-chat window.
• The agent references our knowledge base (FAQs, documentation, and a handful of dynamic data endpoints) to craft a concise, helpful reply in real time.
• If confidence falls below an agreed threshold, the conversation is tagged and transferred to a support rep with the full chat transcript.
• All interactions are logged back into our SaaS dashboard for later analysis.
Preferred stack is flexible—as long as the final solution plugs into the existing React front-end and can be maintained from our Node backend. Integrations with OpenAI or similar LLM APIs, LangChain for retrieval-augmented generation, and vector databases (Pinecone, Supabase, or comparable) are welcome suggestions.
Acceptance criteria
1. Agent answers at least 80 % of common questions in staging tests with no hallucinations of non-existent features.
2. Handoff triggers correctly when confidence dips or when the user explicitly asks for human help.
3. Setup instructions and code are delivered in a clean repo with concise README so my dev team can reproduce the build locally and deploy to production.
If you have shipped a comparable in-chat AI assistant before, especially for SaaS products, your insight will be invaluable as we refine fine-tuning, embeddings, and prompt strategies.