BURAN Guide — Simulated Alexa+ Client Assistant

Inspiration

BURAN is a document-processing platform for law firms. It handles routine extraction of facts, dates and amounts so lawyers can focus on their cases.

Clients also need help using the platform: creating a case, uploading documents, understanding an invoice and finding their completed results. We built the BURAN Guide to provide that assistance within the client's workspace.

What it does

The BURAN Guide is a simulated Alexa+ experience in a web application, built using our own AI agent tools. It is not an integration with the Alexa+ service. The Guide uses records for the selected case to explain: Whether uploaded documents passed the platform's readability checks. What the stored invoice includes and how the price is presented. Whether payment is confirmed, including the distinction between Stripe TEST mode and real payments. The current workflow status and what the client should do next. How to receive the completed delivery package. A delivery-checking capability can compare an uploaded ZIP with invoice promises and produce a report describing the checks and their limitations. The Guide explains platform operations. It does not provide legal advice, pay invoices, change source documents or start processing BURAN existed before the competition. Our hackathon work added and improved the client guidance and delivery-checking experience.

Updates include case-aware guidance, document-status explanations, invoice-based deliverable lists, separate conversation histories for each client and case, history restoration, and protection against showing a previous client's messages or unsent draft after logout.

The demonstration shows the client journey and an actual recorded AI response explaining a €75 invoice and the next step for the selected case.

How we built it

The application uses a Python backend and a browser interface. The live Guide uses OpenAI through a controlled agent tool layer.

There are eleven tools: ten read case records or inspect supplied material, and one saves a delivery-check report. Authentication and case ownership checks restrict access to the client's own records.

BURAN Core provides cryptographic integrity records. These checks can help detect changes to registered records; they do not establish the truth or legal correctness of their contents.

The repository includes source code, setup instructions, synthetic examples and an offline verification route. The scripted TEST provider is deterministic test software, not a live AI mod

Challenges we ran into

Making the assistant useful required more than adding a chat box. We needed to connect explanations to stored case evidence, distinguish unknown states from confirmed events, and keep conversations isolated when users or cases change.

We also corrected an unsent-draft isolation defect and replaced the PyMuPDF dependency with pypdf, then checked reproducibility from extracted packages.

We learned that a helpful assistant needs clear boundaries, reliable context and understandable next steps.

Accomplishments that we're proud of

We demonstrated the live Guide on a synthetic case, checked client and case isolation, preserved existing delivery records, and prepared reproducible offline tests.

Independent reviews covered specific fixes and offline checks. They were not an audit or certification of the entire platform.

What we learned

We learned that a useful AI guide needs more than a chat interface. It must use the selected case's records, distinguish confirmed facts from unknown states, and explain the next step clearly.

We also learned to keep platform processing separate from client guidance, protect conversation history and unsent drafts when users or cases change, and test the application from a clean copy rather than relying only on the developer's environment.

What's next for BURAN Guide — Simulated Alexa+ Client Assistant

We plan to refine onboarding through user testing and explore voice interaction. Voice input and spoken agent responses are future work. All case material shown is synthetic. Payments shown are in Stripe TEST mode

Built With

• css
• html
• javascript
• openai
• pypdf
• python
• stripe