Documents

Documents

To view the documents dashboard, click Sources --> Documents (tab)

Documents as Sources

Click New Source Document to add a training document and title. BotDistrikt parses the document into the following information:

Uploaded Training Document

1. Specifies Document Type.

2. Number of characters in the document.

3. Adjusting the Chunk Size changes the character count in a chunk. You can toggle chunk size from the slider or the Use Recommended button. A large chunk size retains more context while a small chunk size captures more granular semantic data.

4. Chunk Overlap from one chunk to another. You get improved context, a better training model for longer documents, and enhanced coherence. You can increase chunk size if the document consists of numerous unrelated topics.

5. The Output shows the number of responses that will be generated based on the Chunk Size and Chunk Overlap settings.

6. Click Show Example Responses to view sample extracted responses.

7. Click Train AI to train your bot with the document.

The screen redirects to the main Sources (Documents) dashboard.

Uploaded Training Document Dashboard

To ensure the embeddings are created, go to Responses --> Text and confirm whether the embeddings are created.

The Embeddings column displays the LLM name (OpenAI, Vertex AI, etc.). A tick signifies successfully created embedding and a cross signifies no embedding creation for that response. The Tags column indicate the source (document/website)

Successful Embeddings as Response from OpenAI

Troubleshooting:

• If a file is stuck at queued, ensure that your account has not exceeded the respective tier Response Repo limit.

• If a file is stuck at the training phase, click on Resync Source
• If embeddings are not created, ensure that your AI integrated account has not exceeded its token limit.