What happens when a small business customer asks the chatbot a question?

1. The app takes the master microlessons JSON file (the 504 lessons) and splits it into 2,149 chunks.
2. This happens once and is saved, and it only re-runs if the microlessons change.
3. Then it converts each chunk into a vector or an embedding using OpenAI’s text-embedding-3-small mode.
4. These vectors are stored as one matrix that loads into memory when the app starts.
5. When someone asks the Chatbot a question, the question is converted by the app into a vector the same way.
6. Then it is compared against all 2,149 chunk vectors. This takes a few milliseconds.
7. The closest matches are grouped by microlesson.
8. Then the best chunks from the top 2 to 3 lessons are sent to the AI LLM to write the answer.
9. The AI LLM also receives the Question + The persona instructions.
10. The app receives the answer from the AI LLM and finalizes it before sending it to the chatbot.