Show HN: Impersonaid lets you test docs against AI-powered personas

4 days ago 2

Impersonaid helps you evaluate your documentation against LLM-powered user personas to simulate real user interactions and gather feedback, as if it was a UX research session. Of course, it's not real substitute for proper user research, but it can be a useful tool for quick iterations and getting initial insights.

Note that I created Impersonaid using Windsurf and Claude 3.7. This is an experimental tool.

Impersonaid is a tool that helps documentation writers, UX researchers, and product teams evaluate documentation from different user perspectives. By defining user personas with varying levels of expertise, backgrounds, and preferences, you can simulate how different users would interact with your documentation.

# Clone the repository git clone https://github.com/theletterf/impersonaid.git cd llm-docs-persona-simulator # Install dependencies npm install # Make the CLI executable npm link

Copy the example configuration file:

cp config.toml.example config.toml

Edit the config.toml file to add your API keys and customize settings:

[api_keys] openai = "your-openai-api-key" # OpenAI API key anthropic = "your-anthropic-api-key" # Anthropic API key for Claude google = "your-google-api-key" # Google API key for Gemini # Additional configuration options...

Alternatively, you can set the API keys as environment variables:

export OPENAI_API_KEY="your-openai-api-key" export ANTHROPIC_API_KEY="your-anthropic-api-key" export GOOGLE_API_KEY="your-google-api-key"

Impersonaid can be used either through the command line interface or the web interface.

Create a sample persona to get started:

impersonaid create-sample --name expert_developer

This will create a YAML file in the personas directory that you can customize.

Listing available personas

impersonaid list-personas

Test documentation against a persona:

impersonaid simulate \ --persona beginner_developer \ --doc "https://example.com/docs/getting-started" \ --request "What are the first steps to install this product?" \ --model openai

Run an interactive session with a persona:

impersonaid simulate \ --persona beginner_developer \ --doc "https://example.com/docs/getting-started" \ --request "Help me understand this documentation" \ --model claude \ --interactive

Impersonaid also provides a web interface for a more interactive experience:

impersonaid web # or npm run web

This starts a local web server at http://localhost:3000 where you can:

Input documentation via URL or paste markdown content directly
Select from available personas
Choose your preferred LLM provider
Chat with the simulated persona in a user-friendly interface
See responses in a chat-like conversation view

Model-specific limitations

Different LLM providers have varying capabilities when it comes to processing documentation:

Claude: Supports direct URL analysis through prompt engineering. Claude can analyze the content of URLs provided in the prompt.
Gemini: Supports direct URL analysis through its function calling capabilities. Gemini can browse and analyze web content directly.
OpenAI: Does not support direct web browsing. For OpenAI models, the simulator automatically fetches the document content, extracts important sections, compresses it, and includes it in the prompt.
Ollama: Does not support direct web browsing. For local models through Ollama, the simulator fetches the document content, extracts important sections, and compresses it for efficient processing.

Personas are defined in YAML files with the following structure:

name: beginner_developer description: A junior developer who is new to programming and the technology stack. expertise: technical: Beginner domain: Limited tools: Basic understanding of development tools background: education: Computer Science student or bootcamp graduate experience: Less than 1 year of professional experience traits: patience: Low attention_to_detail: Moderate learning_style: Prefers step-by-step tutorials with examples goals: - Understand basic concepts quickly - Find practical examples to learn from - Avoid complex technical jargon preferences: documentation_style: Visual with clear examples communication: Simple and direct explanations

By default, simulation results are saved as Markdown files in the output directory. The files include: