Global AI Milwaukee September Meetup
Details
Agenda Includes:
- Networking / Food
- Brief Introduction / Discuss Group Business
- Featured Speaker
Room: C053/C055
Featured Speaker: Matt Achachlouei
Topic: Beyond "LLM-as-a-Judge": Integrating Deterministic Guardrails & Evidence Graphs
Abstract:
Relying on an LLM to evaluate its own outputs—or using a separate "LLM-as-a-judge"—leaves production systems vulnerable to confident hallucinations. A probabilistic verifier operating in the same semantic space as the generator is fundamentally blind to the exact errors it exists to catch; fluent but incorrect text will often pass validation.
Building defensible, enterprise-grade AI systems requires a shift in focus from semantic plausibility to verifiable, logical correctness.
This session explores how to move beyond prompt engineering by building hybrid architectures that pair generative models with hard-coded, deterministic verifiers. We will dive into the code and system architecture required to ground agentic reasoning in hard evidence rather than probabilistic guesses.
In this session, you will learn how to:
- Construct Evidence Graphs: Trace every material claim and conclusion back to an authoritative source record rather than relying on plausible-sounding model outputs.
- Couple Generative AI with Deterministic Engines: Wire LLMs directly to traditional software constraints, including AST parsers, citation verifiers, and deterministic rules-based logic.
- Preserve Epistemic Distinctions: Classify outputs (e.g., facts vs. sources vs. inferences) to programmatically route claims to the appropriate deterministic check.
Join this session to learn how to enforce correctness through system architecture, ensuring your AI agents are secure, governable, and rigorously engineered.
Featured Sponsors
WCTC [Location Sponsor]
TBD [Food Sponsor]
