Skip to content

Repository files navigation

Open-Bench

Overview

A framework for evaluating LLM-based conversational agents through role-played interactions, using carefully designed personas and questionnaires to assess conversation quality and authenticity.

Agents

Agent A (Interviewer/Host/Salesperson)

  • Leads the conversation by asking questions
  • Supports multiple scenarios:
    • Job interviews
    • Podcast interviews
    • B2B negotiations
  • Maintains conversation flow with:
    • Question quotas
    • Context awareness
    • Topic tracking
  • Uses public information to guide questioning

Agent B (Interviewee/Guest/Buyer)

  • Responds based on detailed persona scripts
  • Follows behavior patterns:
    • Normal: factual responses
    • Slight: minor embellishments
    • Extensive: fabricated stories
  • Maintains consistency with:
    • Previous responses
    • Script boundaries
    • Behavioral patterns
  • Tracks disclosed information

Agent C (Evaluator)

  • Analyzes conversations post-completion
  • Performs two-step evaluation:
    1. Retrieves relevant conversation pieces
    2. Answers questionnaire based on evidence
  • Computes metrics:
    • Answer accuracy
    • Response completeness
    • Behavior pattern recognition

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages