Model Evaluation workshop
RAG Hallucination Control
Measure and cut hallucinations in your RAG system with grounding checks and guards.
- 1 day
- Hands-on in AI Governance Platform
- On your own system
The problem
RAG answers that sound right but are not supported by your documents erode trust fast. This workshop measures faithfulness on your corpus and adds guards that catch unsupported answers.
Who it's for
Teams running a RAG chatbot, copilot, or search assistant over internal documents.
What to bring
Your RAG pipeline (or its logs), a sample of the source corpus, and known bad answers.
The course side
What your team learns
- 1
Diagnose where it breaks
Retrieval misses vs generation drift vs chunking problems, traced on your examples.
- 2
Measure faithfulness
Grounding and claim-level support scoring against retrieved context.
- 3
Layer the guards
Retrieval checks, answer-to-source verification, and refuse-or-escalate rules.
- 4
Cite or abstain
Citation verification and “I don’t know” behavior users can trust.
Builds on these course lectures
- Hallucination Survey
- Risk Analysis of RAG, Prompts, and Agents
- RAG Gains, Fine Tuning Is Rare, and Agents Break Out
The platform side
What your team builds in AI Governance Platform
- Log hallucination as a scored risk on the RAG system
- Record each guard as a control with its threshold
- Track the faithfulness baseline and target over time
Also used: Promptfoo, Guardrails-AI, NLI-style grounding checks
About AI Governance Platform →Your team leaves with
- • A faithfulness baseline for your RAG system
- • A guard design with thresholds
- • A prioritized fix list (retrieval, chunking, prompts)
Book this workshop
Request RAG Hallucination Control
It's pre-selected in the form. Add others to combine them. Contact us for pricing: fill in your details and we'll reply within one business day.