Streamlyft Live Assist - AI Product Knowledge Chatbot
This working prototype gives Customer Success reps at StreamLyft instant, AI-powered answers to product questions in two modes. In Standard Chat, a rep types a question and receives a short, conversational answer drawn from the StreamLyft product guide. In Live Assist mode, the app listens to a live customer call in real time, transcribes the conversation, detects questions as they arise, and proactively surfaces relevant answers on screen - no typing, no putting the customer on hold, no breaking the flow of the call.
The Idea Behind It
The performance consulting analysis behind this case study surfaced a specific finding: 67% of Customer Success reps at StreamLyft could not find reliable, up-to-date product information when customers asked. The internal wiki was distrusted. Product updates arrived through fragmented channels. Reps were being caught off guard on customer calls, and that loss of confidence at a critical moment was contributing directly to early customer churn. This tool was designed to close that gap.
Two modes serve two different use cases. Standard Chat is for preparation and research - before a call, between calls, or when a rep needs to look something up without a customer waiting. Live Assist is for in-the-moment support during an active call. The app listens through the browser microphone, transcribes the conversation in real time, and automatically detects which parts are customer questions versus responses already being given. When a product question is detected, a suggested answer appears in the panel on the right - sourced from the product guide, written in plain language, short enough to read or paraphrase out loud in seconds.
The answer format was a deliberate design choice. Every response is kept concise and conversational - not a documentation extract, but the kind of answer a confident, well-prepared CS rep would give on a call. The goal was a tool that reduces cognitive load in a high-pressure moment, not one that adds to it.
Click the image to access the app 👉
Click here to read the complete Case Study.
Software Used
Claude Code, Claude API (Anthropic), Vercel
Highlights
Two-mode design purpose-built for CS reps: Standard Chat for product lookup and call prep; Live Assist for real-time support during active customer conversations. Both modes draw exclusively from the StreamLyft CS Product Guide.
Live Assist detects questions, not just words: The app analyses the live transcript and identifies when a customer has raised a product question or concern, surfacing a suggested answer only when relevant. The triggering phrase is shown above each answer so the rep knows exactly what prompted it.
Answers written to be read aloud: Short, plain-language, conversational. Not documentation. Not bullet points. The kind of answer a rep can say to a customer without pausing or editing.
Full dual-audio capture on Windows Chrome: The app captures both the rep's microphone and the customer's voice via system audio, so both sides of the call are transcribed. On Mac, microphone-only transcription is supported with a clear on-screen note explaining the difference.
Built in Claude Code, powered by Claude AI, deployed on Vercel: A working, deployable prototype demonstrating real AI integration, not a mockup. Demonstrates how L&D and performance support can move beyond static job aids into tools that work in the flow of work.
