After building the same agent stack twice by hand I wrote a tool to generate it, and deliberately banned the model from writing any of the SQL. Where a language model earns its place in a code generator, and where determinism matters more.
A PL/SQL object with a compile error still gets created, it just gets created INVALID, and most drivers report that as success. Why that surfaces three layers away as a missing-grant error, and the fixed order I now use to trace an agent failure back to its cause.
Registering your own PL/SQL functions as agent tools, and running numpy inside the database through OML4Py. Plus the authentication problem: a stored credential cannot be read back, but an OCI Vault secret can, which removes the hardcoded password entirely.
Building the foundation: the table comments that make natural-language SQL work, the profiles, and the vector index. Includes the policy that catches everyone, since the database needs one of its own and being a tenancy administrator does not cover it, plus why role-based grants silently fail.
Part 1 of a five-part series on building an AI financial analyst inside Oracle Autonomous Database. What Select AI Agent is, why an in-database agent suited this problem better than the managed service, and how the four core objects stack together.
How and why BA Soapbox moved off Wix to a self-hosted Grav instance on Oracle Cloud's Always Free tier, provisioned with Terraform — the stack, the manual steps Terraform did not cover, and the networking, storage and billing surprises underneath it that ate a weekend.
After a three-year hiatus spent relocating back to New York and shifting into a Data Engineer role focused on AI/ML delivery on Oracle Cloud, the blog is back — with a Select AI Agent series already in the works.
A Red Hot session on integration options between Oracle Cloud applications (Fusion, EPM) and the modern data platform on OCI, covering the Lakehouse architecture, extraction methods, and recommended patterns.
A quick survey invite: help build a data set on how people vacation by answering four short questions. Results will be shared once the analysis is complete.
A pandemic-inspired thought piece on why the gap between an event and a company's response — the “recognition lag” — has historically been the biggest bottleneck, and how Augmented Analytics tools like Oracle Analytics Cloud help close it with AI-driven data enrichment, automated narratives, and embedded machine learning.
