Rapid Prototyping: Data extraction using LLMs

At 67 Bricks we use rapid prototyping to validate key architectural choices as early as possible.? A number of? tools, including generative AI coding environments like cursor and windsurf allow us to quickly build interactive prototypes to gather feedback directly from end users, which is invaluable when shaping a new product.

Whilst this approach works well for building new components, it is even more efficient to identify functionality that can be provided by existing products and services, and use low-code tools like n8n and Zapier to stitch these together.? By offloading orchestration tasks to an existing solution, we can demonstrate end to end functionality in a prototype, helping to build customer confidence in the product.

As an example, consider our structured data extraction tool. Whilst this works well for demonstrating single document conversions, its real power lies in applying these conversions at scale.? But how best to demonstrate that? It would be easy enough to script, but a UI driven approach using systems familiar to our users is more compelling.? This is easy to achieve in n8n via a simple workflow:

This processes a google drive folder of PDF files, calls our extraction prototype via a REST interface and stores the results in a Google sheet:

The most effective prototypes are built using interfaces that users can relate to, requiring no additional input (responsibility) from them to carry out the task.? By choosing the most appropriate tool for the job, we can avoid writing boilerplate integration code and focus on the activities which add value to the customer.