Financial Crime
Not Another LLM Talk...Practical Lessons from Building a Real-World Adverse Media Pipeline
An exploration of the core architectural and statistical challenges encountered when scaling a large language model framework into a live financial crime detection system. This session focuses heavily on balancing multi-lingual context windows against API costs and practical throughput restrictions.
AI Signals #27: Not another LLM talk...Practical lessons from building a real-world adverse media pipeline
Presentation delivering a transparent characterisation of moving beyond early-stage proofs of concept to deploying large language models at scale. Drawing on experiences at Comply Advantage processing millions of multi-lingual articles daily, the talk highlights the nuances of parsing long-form text, managing model deprecation cycles, and establishing rigorous evaluation guardrails.
Building a Real-World Adverse Media Pipeline
An overview of practical lessons learned transitioning from LLM proof-of-concept stages to delivering a production-grade, responsibly managed multi-stage financial crime entity extraction pipeline processing over a million news articles a day.