The HR chatbot gives wrong answers
Your company's HR chatbot answers questions about leave, travel and benefits from internal documents, and employees say it's often wrong. Eight parts of its RAG pipeline each have a problem. Pick any part, in any order, and answer its question. When all eight are answered, you'll see which answers employees can trust.
Online 0 / 8Down 0Not checked yet 8XP earned 0
Pick a component
Each component has one question, so read it carefully. Right answers bring it back online.
What users can do
- DownFinds the right policy for each employeeNeeds Documents, Chunker, Embedding model, Vector database, Retriever
- DownGives accurate, grounded answersNeeds Documents, Chunker, Embedding model, Vector database, Retriever, Prompt
- DownLooks up exact codes like TRV-4012Needs Retriever, Vector database
- DownStays safe with untrusted pagesNeeds Chat model
What this incident covers
- Documents (PDFs and wiki pages): Building a Production RAG Pipeline
- Chunker (splits documents): Chunking Strategies
- Embedding model (text to vectors): What are Embeddings?
- Vector database (stores and filters): Vector Databases in 10 Minutes
- Retriever (finds candidates): What is RAG?
- Reranker (orders candidates): Building a Production RAG Pipeline
- Prompt (instructions + context): Evals Basics: Faithfulness & Recall
- Chat model (writes the answer): Defending Against Prompt Injection