Improved genai via Combined Prompting Techniques: Multi-agents (chained prompts), Chain-of-Thought via User Stories and Integration via Function Calling

Larger LLM applications such as database generators present challenges of complexity and quality. Prompt Engineering techniques can help to produce a higher quality LLM application:

• Agents (multiple prompts) can be chained together, like a decision-tree, to manage the complexity and provide an overall structure to the process
• Chain-of-Thought via User Stories: borrowing a concept from Agile software development, asking the LLM to generate User Stories for ‘execution’ in a follow up prompt, can encourage the LLM to think step-by-step, in a Chain-of-Thought like approach that can improve quality
• Integration via Function Calling: integrating the LLM into the greater software system can benefit from Function Calling, where the LLM selects from the available functions to call, and rather than generation application specific data, generates calls to those functions. This helps to integrate the LLM in a loosely coupled manner.

[Image generated by an AI — Adobe Firefly]

Agents

A complex LLM application such as a database generator and modifier can be decomposed for better reliability into multiple Agents. By Agents we essentially mean…