Account-aware agent workflows for advertisers
Applied AI and backend engineering · Production experience
A conversational advertising workflow used Python agents, backend APIs, and ad account operations. A useful result depended on the account data and tools behind the answer.
My contribution
I built the advertising assistant’s request context and tracing feature end to end for production use. The feature assembled context for each request, scoped it to the relevant ad account, and made execution easier to trace.
What it enabled
The workflow could retrieve data for the relevant account. I could see which agent ran, what context it received, and which tools it called from the execution trace.
Why the scale mattered
When a product serves multiple advertisers, an engineer investigating a result needs to know which account supplied the data and how the agent used it. A trace gave me a way to follow the request through the services behind the answer.
That experience informed LLMCheck, an independent tool I built to replay a reviewed case through an application and evaluate the fresh answer.
Implementation details come from engineering notes and code review.