Personas

Overview

Personas represent different types of users and ensure your test data covers diverse perspectives. Generate personas automatically from your knowledge graph to create realistic test scenarios.

Generating Personas

Generate personas automatically from your knowledge graph:

import { generatePersonas } from '@open-evals/generator'

const personas = await generatePersonas(knowledgeGraph, openai.chat('gpt-4o'), {
  count: 5, // Number to generate
  concurrency: 3, // Parallel LLM calls
})

Parameters

Prop

Type

The LLM analyzes your knowledge graph and creates personas with varying:

Expertise Levels - From beginners asking basic questions to experts seeking advanced details

Use Cases - Learning, building applications, troubleshooting, documenting

Communication Styles - Formal technical queries, casual questions, specific requirements

Example personas generated:

;[
  {
    name: 'Junior Developer',
    description:
      'New to TypeScript, learning basic syntax and type system features...',
  },
  {
    name: 'Senior Architect',
    description:
      'Experienced developer evaluating TypeScript for large-scale applications...',
  },
  {
    name: 'Technical Writer',
    description:
      'Creating documentation and tutorials, needs clear explanations...',
  },
]

Each generated test sample is associated with a persona, ensuring questions reflect realistic user scenarios.

Custom Personas

Define custom personas for specific test scenarios:

import { persona } from '@open-evals/generator'

const customPersonas = [
  persona({
    name: 'Security Auditor',
    description: 'Focused on security implications and best practices...',
  }),
  persona({
    name: 'Performance Engineer',
    description: 'Concerned with runtime performance and optimization...',
  }),
]

const samples = await synthesize({
  graph,
  synthesizers,
  personas: customPersonas, // Use custom personas
  count: 100,
})

Mixing Generated and Manual Personas

Combine automatically generated personas with custom ones:

const generatedPersonas = await generatePersonas(graph, llm, {
  count: 3,
})

const allPersonas = [
  ...generatedPersonas,
  { name: 'Compliance Officer', description: '...' },
  { name: 'DevOps Engineer', description: '...' },
]

const samples = await synthesize({
  graph,
  synthesizers,
  personas: allPersonas,
  count: 100,
})