Request a Call
Spinner

Processing...

Business Overview

Our client, an ambitious startup, has embarked on a mission to streamline the book review and suitability assessment process for children’s literature, catering to the needs of publishers, bookstores, and educational institutions. Their core idea centers on developing a comprehensive AI-powered application designed to automatically and rapidly review, rate, and summarize books. Because the client is a member of the AWS Partner Network (APN), it allows them to gain significant support and funding from AWS, which was supposed to increase usage of AWS cloud services.

With our decades-long partnership with AWS and our seasoned and experienced team of professionals, the client approached us to transform their vision into a market-ready solution. Our goal was to provide end-to-end development services, building a robust and scalable cloud-native application from scratch.

1_600x405@2x

Challenge

The primary challenge was the inherent technical limitations of large language models (LLMs). These models possess a window limit, meaning they can only process a finite amount of text at one time. This capacity was significantly less than the content of even a single, moderately sized book. Overcoming this constraint by enhancing the LLMs would have required substantial additional hardware—specifically GPUs and TPUs—which would have drastically inflated project costs and thus wasn’t a viable option.

Solution

We started with the implementation of an advanced Amazon Nova-based solution, leveraging both its text-based and multimodal capabilities for efficient content evaluation. Additionally, we integrated Amazon Nova multimodal capabilities for comprehensive image review, particularly crucial for assessing graphic novels. The final application is designed to accurately identify and flag restricted content within books, such as sexually explicit content, hateful speech, violence, and drug use.

2_600x405@2x

To overcome the inherent window limit of LLMs without incurring additional hardware costs, we devised a highly effective and cost-efficient custom solution: chunk-based processing. This method breaks down an entire book into manageable segments that the LLMs can process within their text limitations. The model then meticulously scans each chunk for restricted content, classifies it, and subsequently compresses and hierarchically summarizes the plot, providing an overall book score.

3_600x405@2x

Furthermore, to ensure a cost-effective and rapid solution for the client, we implemented parallel processing in batches. This significantly speeds up book processing times and reduces the overall operational costs of the application.

Key features of the application:

  1. Automatic detection and marking of restricted content within books

  2. Extracting captions and analyzing illustrations for restricted content

  3. Comprehensive book summarization based on both text and image analysis

  4. Specifying a minimum age for the book

  5. Scoring books according to US laws regarding content suitability

Our team continues to expand the app’s features. In the next update, we plan to roll out several new functionalities, including:

  1. Efficient bulk upload and batch processing for multiple files

  2. Document-type dependent image processing

  3. User feedback mechanisms for fine-tuning book scores

  4. Multilanguage support

  5. UI display improvements to clearly show flagged images

  6. Retry mechanism implementation to enhance product scalability for handling large texts

  7. Navigation enhancements to integrate document readers for rapid content validation

Outcome

Our collaboration culminated in the successful launch of a market-ready product that immediately captivated the industry and attracted substantial additional investments for the application enhancements and expansion. This success has fueled an exponential growth trajectory, with the product now on a path to scale rapidly thanks to strong investor confidence.

1000+

active users

1026

books processed to date

5-minute

average processing time for medium-sized books

3000-page

book processing capability

Team:

Team:

7 Experts ( Project Manager, 2 Data Scientists, Data Engineer, DevOps, NodeJS developer, Front-end developer )
Tech stack:

Tech stack:

AWS Lambda & CLI, AWS Bedrock, LangChain, Anthropic Claude, Amazon Nova, Amazon Step Functions

REQUEST A CONSULTATION

Contact us   

Relevant Case Studies

View all case studies

AI Integration: Streamlining Accreditation Workflow in Education

Education

Success Story AI Integration: Streamlining Accreditation Workflow in Education  image

Infosec: Migration of an Education Platform to CMS Optimizely

Education

Success Story Infosec: Migration of an Education Platform to CMS Optimizely image

MindTap

Education

Success Story MindTap image

Next-Gen eLearning Platform for Medical Schools

Healthcare

Education

Success Story Next-Gen eLearning Platform for Medical Schools image

LibraryPass Cloud Infrastructure: Cost Optimization with AWS

Publishing

Education

Success Story LibraryPass Cloud Infrastructure: Cost Optimization with AWS  image

Pharmaceutical Software Modernization and AI Implementation

Healthcare

Success Story Pharmaceutical Software Modernization and AI Implementation image
01

Contact Us

Accessibility Adjustments
Adjust Background Colors
Adjust Text Colors