Document Intelligence
Extracts the text and tables from uploaded PDF and Word files, including scanned pages, and splits them into passages.
Built on Amazon Textract
Technology
The Technology Behind Cited Answers
Brevton combines document text extraction, semantic search and large language models to answer questions about your own papers, patents and reports — and show which source each point came from. It's being built on AWS.
Early access·No credit card required
For every question
- 1Find the relevant passages
- 2Write an answer from them
- 3Attach the sources
The Stack
Eight Parts, One Rule
Answers should come from your documents and point back to them. Each part below handles one step of that, from reading a file to citing it.
Language Processing
Picks out entities and key terms and groups documents by topic — the groundwork for search, clustering and citations.
Semantic Search
Each passage becomes a vector embedding, so a question about “capacity fade” can find a paper that only says “cell degradation”.
Built on Amazon OpenSearch
Large Language Models
Read the passages retrieved from your library and write answers, summaries and reports from them — not from general knowledge.
Built on Amazon Bedrock
Source Citations
Every passage keeps a reference to its document, so each point in an answer, summary or report links back to its source.
Research Analytics
Looks across a whole project to show which topics come up most, how documents cluster and where coverage is thin.
Automation
Handles the repetitive steps of a review: paper-by-paper summaries, sorting by topic, pulling out references and report outlines.
Built on AWS Lambda
Security & Privacy
Managed sign-in, HTTPS on every page and upload, and documents visible only to the workspace members you invite.
Built on Amazon Cognito
Cloud Infrastructure
How Brevton Is Built on AWS
Storage, search, sign-in and model access run on AWS managed services, so our small team can spend its time on the product itself.
This is the architecture we're building during early access. Specific services may change as the product develops.
How we protect your documentsAmazon S3
Storage for the documents you upload to a workspace.
Amazon Textract
Extracting text and tables from uploaded PDFs, including scanned pages.
Amazon OpenSearch
Vector search over document passages, so questions match by meaning.
Amazon Bedrock
Foundation models that write answers, summaries and reports from retrieved passages.
AWS Lambda
Serverless functions for document processing and the application's APIs.
Amazon Cognito
User sign-in and workspace access.
Amazon SES
Transactional email, such as sign-in messages and workspace invitations.
Amplify & CloudFront
Hosting and content delivery for the Brevton web app.
Design Principles
The Rules Behind the Stack
Six principles we follow when deciding how Brevton should work and what to build it on.
Grounded in Your Documents
Answers are written from passages retrieved from your library, not from the open web.
Show the Source
Every answer lists the documents it drew on, so you can check each point yourself.
Private Workspaces
Documents belong to a workspace and are visible only to the members you invite.
Managed AWS Services
Storage, search, sign-in and models run on AWS managed services rather than servers we run ourselves.
Serverless by Design
Document processing and APIs run on AWS Lambda, so capacity follows usage instead of fixed servers.
Room to Change Models
Amazon Bedrock offers several foundation models, so we can adopt better ones without rebuilding the product.
Early Access
See It Work on Your Own Documents
Request access to try Brevton with your own papers and reports, or book a demo and we'll walk you through how search and citations work.
Early access·No credit card required