Knowledge agents: answers from your documents, cited

Answers from your policies, manuals and PDFs, including the images and diagrams inside them.

Cites the source document and page with every answer.
Reads the images and diagrams inside your documents, and shows the relevant figure, not just text.
Agentic retrieval checks whether what it found actually answers the question, and searches again if not, instead of guessing.
Available in Microsoft Teams. Slack available on request.
Runs on Azure AI Search, or on a self-hosted vector database when your data needs to stay on your own servers.

How it works

Step 1:
Question asked in Microsoft Teams
Step 2:
Agent retrieves from your documents, including images and diagrams
Step 3:
Grades whether it answers the question
retries capped
no
yes
Step 4:
If not, rewrites the search and retrieves again
Step 5:
Answers with the source, page and figure
At step 3: if yes, go to step 5; if no, go to step 4.
Step 4 goes back to step 2 (retries capped).

Example flow. Each agent is built per project.

What we've built

A multimodal document assistant in Microsoft Teams, built on Copilot Studio, Azure AI Search and Azure Functions. It answers from PDFs using section-aware retrieval and shows the relevant diagrams and images inline in Teams, served through short-lived, user-delegation SAS links.

A self-correcting retrieval system built in LangGraph. It grades the documents it retrieves, rewrites the question when results are weak, caps its retries, and keeps track of the conversation across follow-up questions.

Contact Us

Have a question? Send us a message and we'll get back to you within 24 hours.