Plan as Narrative, Test as Code

Leverage an LLM to make community planning better.

Objective

Here I introduce a new approach to making a plan. The approach leverages the new advances in Large Language Models (LLMs) to translate a narrative chain of dependencies into code. The approach also highlights my prompt-engineering contribution to the LangChain open-source libray called the Causal Program-aided Language (CPAL) LLM chain. This implementation of CPAL is purely conceptual.

Usage

• User A writes causal narrative to define a plan or more formally a work-breakdown-structure (WBS).
• The LLM translates User A's causal narrative of the plan into code.
• A teammate User B writes a hypothetical question of her speculated change to the original plan.
• The LLM translates User B's question into a query.
• The application runs the query and generates a report on the impact of User B's speculated plan change on the plan's outcomes.

Research questions

• Is there public data to make a prototype of this concept app?
• How can we add time as a parameter for each work span, and total time as an outcome?
• How can we add cyclic dependencies?
• What optimization code already exists to help a planner?
• Geo-spatial queries and impact analysis?
• Time and cost and ROI optimizations?

References

• Extend the my CPAL experimental work in LangChain, LangChain PR here.
• LangChain's tweet on the CPAL is here

Cool stuff by others

• 2023 AutoScrum™: Automating Project Planning Using Large Language Model Programs