I would use whichever is best for the company, compatible with your long-term vision, and fits within your price range. There is BigQuery, Snowflake, Databricks, and the like which work well.
Hey Brock, bringing the data to claude is one thing, any thoughts on the harness around the model? Skills files that recursively improve, tool calls that write back to source systems, etc
Thomas, (sorry, not Brock, but also a CPA using these tools and love Brock's posts!) ,
I have seen some folks build a "memory" skill into prompts when run, where it's baked in to record anything learned from a run and record it into a memory file read at the beginning of each run. Also works when troubleshooting deliverables as well. This memory file could live in the data lake, or elsewhere.
Leads to some token inflation on each run, but has been the best solution I've seen just yet.
Thanks for your comment, Thomas! What Baron said is what I’ve seen most people do, as well. I’ll do more research and see what else I can come up with in future writing.
Nicely done, Brock! Great stuff!
Thanks, Christian!
Brock, Great article and nicely done! Where do you organize the semantic layer and data warehouse in your experience?
Appreciate the support! Do you mean which platforms or how do you set them up?
What are you using for a data warehouse/lake?
I would use whichever is best for the company, compatible with your long-term vision, and fits within your price range. There is BigQuery, Snowflake, Databricks, and the like which work well.
Hey Brock, bringing the data to claude is one thing, any thoughts on the harness around the model? Skills files that recursively improve, tool calls that write back to source systems, etc
Thomas, (sorry, not Brock, but also a CPA using these tools and love Brock's posts!) ,
I have seen some folks build a "memory" skill into prompts when run, where it's baked in to record anything learned from a run and record it into a memory file read at the beginning of each run. Also works when troubleshooting deliverables as well. This memory file could live in the data lake, or elsewhere.
Leads to some token inflation on each run, but has been the best solution I've seen just yet.
Appreciate your support, my friend. 🙏
Thanks for your comment, Thomas! What Baron said is what I’ve seen most people do, as well. I’ll do more research and see what else I can come up with in future writing.