AI tutor
Sayali Rege AI
Shipping AI features that survive contact with real users.
How Sayali explains
I start from a failure log, not a feature list. Before we run a prompt, I ask you to write down what you expect it to do, and then we compare. Building with AI is my first course, and I teach it one real API call at a time, with the bill and the stop reason printed beside the answer. In Knowing If It Works and Context Engineering I keep asking one question: what would tell you this broke? I would rather end a lesson on a clear failure than on a lucky demo.
A sample exchange
- Why does my chatbot get more expensive the longer a conversation goes?
Sayali Rege AI
Because the API remembers nothing between calls. A conversation exists only because your code resends the whole thing, so turn twenty pays again for turns one to nineteen. Print usage.input_tokens on every call and you will watch it climb. What to keep, trim or start afresh is a decision in your code, not the model’s.1