Start with a shared data contract.
I worked across teams to define common AI delivery metrics. Consistent contracts helped turn different implementations into useful reporting without pretending every team worked the same way.
20+Teams / business units in scope
5Highest-spending orgs in discovery
Start with the
actual problem.
AI activity was spread across teams with different pipelines, workloads, and definitions of success. Cost alone could not tell the full story. Teams needed useful signals about usage and output quality, and leadership needed a shared view that made those signals comparable.
The work behind
the outcome.
Connect usage to quality.
I built Python telemetry ingestion and LLM payload parsing for token usage, latency, and throughput. Quality and drift flagging, together with human-review tooling, added context to the cost picture.
Meet each team where it already worked.
Discovery and governance involved the five highest-spending organizations. Integration wrappers and middleware addressed differences in schemas, rate limits, and regional compute boundaries.
How the pieces connect.
01Team-specific pipelines
02Common data contracts
03Usage + quality signals
04Flags + human review
05Director-level reporting
Explanatory system view · reconstructed from the project
Hands on.
People with me.
I contributed the ingestion, parsing, flagging, and review tooling, including a React reporting interface. I also worked with infrastructure and partner teams to resolve integration issues and make the metrics useful across organizational boundaries.
PythonLLM telemetryScikit-learnReactData contractsHuman review
What came out of it.
The work spanned 20+ teams and business units, with focused discovery and governance involving the five highest-spending internal organizations. It brought resource tracking and quality signals into reporting that could inform decisions about AI deployments.
CONTEXT & LIMITS
The organizational scope describes the breadth of the work. It is not a count of independently verified services or deployments, and no specific dollar savings are claimed here. The internship ended in August 2026.
Lori Comunica
A better way to be understood.