Beyond the hype: practical AI for enterprise engineering with Disleve Kanku

Where can AI reduce toil without removing ownership — and why does that question start with the data, not the model?

In this session, Disleve Kanku — Data Engineer II @ Dana-Farber Cancer Institute and Founder & CEO of OncoSys AI — brings a data engineering lens to network operations, arguing that AI systems succeed or fail based on problem definition and data quality long before the model matters.

He covers:

• The five-layer framework for any AI system — problem, data, model, workflow, and outcome — and why a better model rarely rescues a system when one of these layers is weak
• Why “trust” has to be built into the pipeline: verifying that data isn’t silently altered or hallucinated on as it moves through AI systems
• Lessons from building OncoSys AI, a data-readiness and trust-layer platform born out of research on messy healthcare data
• A practical taxonomy of AI tooling — deterministic automation, grounded co-pilots, and controlled multi-agent systems — matched to the right use case rather than defaulting to the most powerful model
• The five levels of agentic maturity: search, summarize, recommend, plan, and execute — and where human approval should sit in that progression
• Why context control (role-based access for agents) and action control (evaluation, approval, and audit agents) are essential guardrails as multi-agent systems take on more autonomy

The session reframes the AI conversation around a simple discipline: define the problem, get the data right, and keep a human in the loop — because no amount of model sophistication compensates for skipping those steps.