User Guide
- Manage Prompts
- Create Prompt
- Edit Version
- Generate Version
- Rename Version
- Delete Version
- Set Primary Version
- Lock Version
- Compare Versions
- Token Count and Cost
- Update Prompt Use Case
- Copy Prompt to Workspace
- Move Prompt to Workspace
- Test Prompts
- Prompt Chat
- Response Alignment
- Optimize Prompts
- Prompt Scoring
- Scoring Criteria
- Prompt Heatmap
- Prompt Balance
- Prompt Risk
- Workspaces
- Add Workspace
- Rename Workspace
- Delete Workspace
- Switch Workspace
- Projects
- Add Project
- Delete Project
- Update Project Brief
- Assign Prompt to Project
- API Clients
- Add API Client
- Edit API Client
- Delete API Client
- Regenerate API Client
- Require API Client
- Connect AI Providers
- AI21 Labs
- Amazon Bedrock
- Anthropic
- Azure OpenAI
- Azure Serverless
- Cohere
- Deep Infra
- Fireworks AI
- Google Gemini
- Groq
- Hugging Face
- Hyperbolic
- IBM WatsonX
- Inception
- Meta
- Mistral
- Moonshot AI
- OpenAI
- Perplexity
- Together AI
- xAI
Response Alignment
On this page you'll find the following:
Get Started
Response alignment uses the LLM-as-judge technique to determine how well AI responses align with the intent and expectations of the prompt, use case, project brief (if applicable), and the user query.
To get started, go to the version's Chat tab. Use the input box to enter at least 1 query or question to generate an AI response or series of responses.
Tip: Response alignment will analyze all AI responses in the chat session.
Analyze Responses
To analyze responses for alignment, click the purple Analyze responses button in the right sidebar. This will bring up the Analyzing Responses modal box.
After a few moments, the overall alignment score will populate in the right sidebar, along with the alignment scores for each of the AI responses in the chat session.
Clicking the alignment score for each response will bring up a modal box with detailed analysis that includes a summary and list of pros and cons for how well (or not) the response aligns with the prompt, use case, project brief (if applicable), and the user query.
