Recordings coming soon!
Talk 1: Processing Thousands of Sensitive Documents for Retrieval
Speaker: Brandon Lim, Senior ML Engineer @ Faculty
Abstract: Most document processing pipelines look straightforward until you throw thousands of real-world files at them. Inconsistent formats, sensitive data, access controls, and scale all expose the gaps that prototype pipelines paper over. This talk covers the full journey from raw document ingestion to production-ready AI search: including data storage strategies, embedding and indexing, retrieval architecture, and how to keep quality and compliance intact as your corpus grows.
Key Takeaways: Participants will leave with a practical understanding of how to design document pipelines that scale.
Talk 2: Guardrailing LLMs in High-Stakes Workflows
Speaker: Prudence Leung, Senior Data Scientist @ Faculty
Abstract: In high-stakes corporate workflows, production reliability is on the frontlines of preventing severe financial, legal or regulatory risk. It’s about architecture: validating outputs at every boundary and orchestrating multi-step pipelines so they’re resumable, observable, and safe to retry.
This talk covers practical layers of that architecture, such as using Pydantic to turn LLM outputs into hard pass/fail gates, and durable execution frameworks like DBOS and Temporal to make multi-step pipelines resumable and safe to retry. The goal throughout is keep deterministic logic deterministic, and contain the probabilistic parts.
Key Takeaways: Participants will leave with concrete patterns for validating LLM outputs in production and a practical model for orchestrating reliable, auditable LLM-powered workflows.
Talk 3: Panel – The Autonomous Underwriter: How do you integrate GenAI agents into legacy insurance systems without breaking the business?
Panellists: Chris Mullan, Head of Data and AI @ CFC Underwriting, Divish Rengasamy, Lead ML Engineer @ Faculty, Hanne Carlsson, Senior Data Scientist @ Faculty. Hosted by Michelle Conway, Lead ML & AI Engineer @ Lloyds Banking Group.
Abstract: Agentic AI promises to transform underwriting but legacy systems, regulatory constraints, and organizational complexity make the path from demo to production far harder than it appears. Beyond the software engineering, this is a massive change management and operational transformation. Drawing on real-world experience, this panel unpacks the friction of decoupling agents from brittle dependencies, managing risk acceptance, and building systems that improve over time without creating new operational risks.
Key Takeaways: An honest assessment of where agentic underwriting works today and where it fails. Practical frameworks for bridging the underwriter expectation gap and establishing true business ownership.
Supported by
