Episode Details

Daniel's habit of manually starting new ChatGPT threads is actually smart — but it doesn't have to be manual. This episode walks through exactly how to build a lightweight classifier that detects topic shifts and spawns fresh threads automatically, plus the more ambitious version that routes messages to whichever past thread they belong to. We cover the engineering tradeoffs, cost analysis (spoiler: pennies per day), and how to implement it with LangChain in about fifty lines of Python. If you've ever wondered why large context windows don't solve the real problem, or how to build a single continuous conversation that keeps context clean behind the scenes, this one's for you.