Automating AI Decisions: System One Classifiers (Jev) and LLMs in Practice Training Course

1. System One vs. LLMs in Decision-Making
• Characteristics of System One models and their place alongside Large Language Models (LLMs)
• The Jev model (TypeSafe) and its application in text classification
• Comparison of System One and LLMs: cost, speed, and prediction accuracy
• Classic classification tasks in practice: spam detection, sentiment analysis, and more

2. Initial Candidate Screening in Recruitment – A System Supporting HR Professionals
• Extracting structured data from CVs: work experience, skills, education, and language proficiency
• Evaluating candidate profile fit against job descriptions and assigning decisions: rejection / further verification / interview invitation
• Building a complete application used daily by recruiters
• Fast classifier: System One for initial screening, LLM for formulating justifications and proposing interview questions

3. Intelligent Customer Support – Automated Ticket Routing and Prioritization
• Recognizing ticket topics and forwarding them to the appropriate team
• Detecting unwanted content and moderating incoming messages
• Determining sentiment and urgency, establishing service order, and escalating complex cases to agents
• A project where System One handles rapid classification, while the LLM generates responses only when needed

4. Model Selection for Queries – An AI Assistant with Intelligent Routing to Specific Models
• Assessing the complexity of user questions (direct answer / accessing knowledge / requiring tools)
• Routing queries to the appropriate model: cheap and fast for simple tasks, advanced LLMs with tools (search engines, knowledge bases, APIs) for complex ones
• Handling uncertain cases: when the classifier has low confidence, the query is passed to a more powerful model
• Breaking down complex questions into smaller steps routed to different models