Generative AI for Genealogy – Part VI

Follow‑Up Questions: The Real Fun Begins

A follow‑up question means repeating the entire dance, but with more context:

• Prompt
• Question 1
• Tool call
• Data
• Answer 1
• Question 2
• …

Let’s visualise it:

1. User types question
2. App determines skill + examples, sends prompt + question
3. LLM asks for a tool
4. App sends everything + tool call + data
5. LLM replies with an answer
6. App displays answer
7. User asks follow‑up
8. App sends new prompt + full conversation + new question
9. LLM asks for a tool
10. App sends everything + tool call + data
11. LLM replies
12. App displays answer

And yes, we skipped the part where a different LLM determines the skill. Let’s add that back in:

Chaining Questions and Answers

We’re not limited to a single tool call per question. The LLM can ask for multiple tools in one go.

Example:

“How old are Bart and Lisa?”

LLM replies:

``` get-data:Bart```
``` get-data:Lisa```

You return the data, and it answers:

“Bart is 10 years old and Lisa is 8 years old.”

Brilliant.

But some questions require sequential reasoning.

Example:

“Where was Dave born? And which counties neighbour that place?”

To answer:

1. First find Dave’s birthplace
2. Then find neighbouring counties

Models like Reasoner v1 can handle this, but you need two mechanisms:

• If the LLM asks for a tool, you provide the data and ask again
• You detect FINAL ANSWER: to know when it’s done

This loop continues until the LLM stops asking for tools and finally answers the question.