Skip to content

Source-Grounded Business Summaries

SageMaker models SageMaker model

Create review-required draft summaries and action candidates from English, Spanish and Portuguese business text. Version 0.1.0 includes the complete selected source segment and Unicode offsets for each item.

Delivery
SageMaker model package in your own AWS account

Review workflow

Use meeting notes, project updates or vendor correspondence that you are authorized to process. The language model selects segment IDs; the server copies their exact text. Literal source location is checked, while semantic entailment and summary factuality still require your review. Owner and deadline remain null unless supported by an explicit named person and valid literal ISO date.

The package uses pinned original Qwen weights without additional Software Sushi training. It accepts bounded text only, runs offline with network isolation and does not create tasks, send messages or operate tools.

Deployment and billing

Qualified types are ml.g6e.4xlarge for real-time and ml.g6e.2xlarge for batch in us-east-1, each with one 48 GB GPU. Software is USD 0.50 per instance-hour, with a 14-day software trial for eligible buyers. AWS GPU infrastructure and other AWS services are separate, including during trial. Confirm quotas, availability and the active offer before deployment.

Walkthrough

Full walkthrough · Vertical Short. The illustrative interface replays actual certified AWS outputs for fictional development examples; it is not an included application. English narration is AI-generated.

Next steps

Upstream project

The model is based on Qwen/Qwen3.8-27B-FP8. Read the source/license notices and model card for the exact upstream identity, dependencies and evaluation scope.

Last updated November 15, 2026 · Suggest an improvement