Tamara DinneenGet in touch

Enterprise AI, built in public

AI that earns
trust in banking.

For enterprise AI deployments that institutions can trust.

About my work

About

I make the case for trusting a system, not just using it.

I'm Tamara Dinneen. I spent two decades inside enterprise software at SAP, Telstra and Oracle, on the commercial side: running global partner marketing, channel economics and reporting at scale across dozens of countries and languages. Then I stopped commissioning AI systems and started building them myself.

Today I architect and ship production AI at Oracle: the global system that resolves partner entities with machine learning and turns them into executive reporting, and a sales-enablement engine that surfaces the right proof point from thousands of customer stories in real time. Alongside it I took a Masters in ML and AI and am partway through a doctorate. My thesis re-engineered a leading protein-design model and beat the published benchmark on every single test case.

I build AI that holds up under scrutiny: systems that cite their sources, know when to abstain, and stay inside a cost ceiling set before the first line of code. I work the same way across sectors, and I document it in the open, so the people relying on the result can see exactly how it was reached.

Contact meDownload CVFood detection demo

Across industries

01 / 04

Banking

Traceable AI for documents that cannot afford to be wrong.

See the finance practice →

02 / 04

Marketing

Cost-controlled AI for content and analytics at real volume.

On request →

03 / 04

Logistics

Decision support that stays legible to the people accountable for it.

On request →

04 / 04

Life Sciences

Citation-grade AI for scientific literature and regulatory text.

On request →

Core expertise

Trustworthy retrieval
Systems that ground every answer in a real source, cite where it came from, and abstain when the evidence is thin. Hybrid keyword and semantic retrieval, reranking, and grounded generation, so the output can be checked rather than taken on faith.
Cost-controlled deployment
AI that stays inside a budget set in advance. Token economics modelled before launch, cheaper models used where they suffice and stronger ones reserved for where they earn their cost, so spend is a decision rather than a surprise on the invoice.
Agentic architecture
Multi-step, multi-agent systems that stay legible. Each step is inspectable and accountable, with clear handoffs and guardrails, so a complex pipeline does not become a black box that no one can reason about.
Evaluation and rigour
Pre-registered methods, contamination testing, calibrated confidence, and limitations stated plainly. The discipline that lets a result hold up under scrutiny from the people who have to rely on it.