App Overview
The App consists of the following pages:
- Chats
- Insights
- Datasets
Chats
The Chats page displays raw chats with their respective feedback.
Datatable includes:
- User Query
- LLM Response
- Satisfaction Score
- Date and Time
- Version
Clicking on a chat reveals:
- Feedback:
- Gesture (thumbs up/down or rating)
- Message
- Origin (source)
Insights
Issues
The Issues page shows chats clustered into similar Issues, with scores and analytics to help prioritize.
Analytics includes:
- Number of issues over time
- Sentiment of user queries
- Impact score over time
More Insights Coming Soon!
Datatable includes:
- Issue Name
- Description
- Impact Score
- Number of Chats
- Time of Creation
- Issue ID
Clicking on an issue shows a datatable of associated chats, including:
- User query
- AI response
- Feedback Gesture (if any)
- Date and time
- Version
Clicking on a chat within an issue reveals:
- Feedback:
- Gesture (thumbs up/down or rating)
- Message
- Origin (source)
- Root Cause Analysis (RCA):
- Knowledge hole probability (the probability the given chat issue is due to missing context)
- More coming soon
- Evaluation metrics:
- Query Sentiment
- Response Sentiment
- Response Relevance
- Faithfulness
More Insights Coming Soon!
Topics
The Topics page shows chats clustered into similar topics to help prioritize easily.
Analytics includes:
- Topics over time
Datatable includes:
- Topic Name
- Description
- Number of Chats
- Time of Creation
- Topic ID
Global Actions
The Global Actions is available on all pages and allows you to:
- Search the primary column (specific to the current page)
- Download the current datatable in .csv format
- Filter by:
- Date range (preset or custom ranges)
- Impact Score (0-10 slider; only available on Insights page)
- Custom rules coming soon!
Datasets
Create Datasets
Create Datasets is available on all chats including issues and topics. It allows user to create a dataset with specific queries and repsonses. This datasets can be edited further by domain experts (coming soon).
Upload Datasets
Upload Datasets allows user to upload a dataset with .CSV format. The structure of the file should be the following (query, response, date (optional), expert_response (optional), context(optional)).
Generate Synthetic Ground Truth
Insights layer can generate ground truth reponses if there is no knowledge hole and the user is not satisified by LLM response. The evaluation metrics are being calculated using the generated ground truth with versioning (coming soon).