...rather than iterating on blueprints for months.
Tangible results are 1000 times more relevant than brainstorming on Zoom for hours.
When it comes to prototyping LLms, what usually works is to follow this process:
Create a Proof of Concept promptly.
Develop a system for improvement, outlining testing methods and evaluation processes that are understandable to all stakeholders.
Begin small-scale iterations.
Progress to a Minimum Viable Product once stability is achieved.
Add new elements, then evaluate and iterate, scaling up until a milestone satisfactory for an MVP is reached.
Transition to a large-scale or industrial approach.
This method reflects successful change management execution.
When you have stakeholders who are globally distributed and have expertise in various business domains beyond IT/Engineering, frequent workshops are needed to ensure you are moving in the right direction. These workshops present results of iterations, supporting data, and insights into potential integration challenges.
I used "fractal" because achieving a precise implementation of a product from scratch is never feasible.
Each new iteration can bring forth ideas that may surpass the original intent, as people now have a tangible reference point for discussion and, more importantly, decision-making.