Testing Recurring Payment Retry Guidance Against the Documented Workflow
An anonymized investigation of an AI support assistant that repeatedly contradicted documented recurring-payment retry behavior.
Investigations & writing
AI support reliability, billing workflows and the systems behind customer guidance.
Building Regression Tests for an AI Support Workflow
How I designed and ran a regression suite for a billing support workflow in Fin, identified a repeated failure pattern and validated a targeted knowledge change.
When Query Rewrite Changes the User's Question in a RAG Pipeline
A Quivr RAG investigation into semantic drift during query rewrite, model-dependent behaviour and why stronger prompts did not reliably preserve user intent.
Testing Help-Center Retrieval Before Generation
A small retrieval evaluation project comparing TF-IDF, local embeddings and hybrid retrieval on a synthetic SaaS help-center dataset.
Designing a UI QA Inspector That Separates Screenshot Evidence from DOM Evidence
An AI-assisted frontend QA experiment about separating what a screenshot shows from what the DOM proves.
Designing a Review Queue for Revenue Risk Support Messages
A small AI/frontend project about turning messy support messages into reviewable revenue risk signals.
Why SaaS Billing Systems Still Get Failed Payments Wrong
A teardown of the flat retry pattern in SaaS billing systems and how treating all failures the same can create missed recovery and access problems.
Working on a similar problem?
Tell me which support workflow you need to investigate or make repeatable.
Discuss your workflow