This talk presents a detailed case study of implementing agentic AI workflows in production data systems at AstraZeneca, one of the world’s largest pharmaceutical companies.

THE CHALLENGE:
AstraZeneca’s data platform processes petabytes of pharmaceutical data across thousands of pipelines. Traditional monitoring generated alert fatigue, with teams spending 40% of their time on reactive incident response. Schema changes caused cascading failures. Mean time to recovery (MTTR) averaged 4+ hours.

THE IMPLEMENTATION:
Yusuf Ganiyu led the implementation of agentic AI systems that autonomously:

• Investigate data quality anomalies, reducing investigation time by 70%
• Detect and adapt to schema changes before failures occur
• Diagnose root causes and suggest remediations
• Execute approved fixes with human-in-the-loop governance

ARCHITECTURE DETAILS:
The talk covers the complete technical implementation:

• Large language model (LLM) integration patterns with existing Airflow orchestration
• Tool design enabling agents to query metadata, run validations, and execute fixes
• Guardrails to prevent hallucination in data-critical, pharmaceutical contexts
• Cost management practices achieving an 85% reduction compared to initial projections

MEASURABLE RESULTS:

• 70% reduction in anomaly investigation time
• 60% decrease in schema-related incidents
• MTTR improved from 4+ hours to under 45 minutes
• Zero compliance violations from agent actions in a regulated industry

Attendees will receive a reference architecture and an adoption roadmap template. This session is led by Yusuf Ganiyu, who brings hands-on experience in deploying production-grade agentic AI systems in highly regulated enterprise environments.