AWS describes the Contract Intelligence Platform architecture, which replaces RAG for aggregation queries over contracts
AWS has published a case study of a platform architecture for analyzing vendor contracts: instead of classic RAG, it uses AI agents for structured extraction of data into a database, because RAG fails at questions requiring sums and comparisons across hundreds of documents.
AWS described in a technical blog post the architecture of a platform for analyzing vendor contracts called the Contract Intelligence Platform. The starting problem is that common tools based on Retrieval Augmented Generation (RAG) work well for questions about a single document, but fail at aggregation questions across an entire contract portfolio (for example, the total value of all contracts or an overview of upcoming expirations), because RAG always returns only a limited number of the most relevant text snippets and cannot sum or compare data across the entire dataset.
According to AWS, the solution is structured extraction: AI agents pull key data from individual contracts into a database that can then be queried with analytical tools, while the original documents remain in a knowledge base for queries about individual contracts. The platform is built as a React web application; the extraction and verification agents are created using the Strands Agent SDK and run on the Amazon Bedrock AgentCore runtime, which provides serverless hosting, automatic scaling, and session isolation. To verify accuracy, a combination of two different models is used, and Amazon Textract resolves any discrepancies regarding signatures. Access to data is governed by Cedar-based security policies within Bedrock AgentCore. Users then view the data through built-in dashboards (Amazon QuickSight) and a natural language chat.
As an illustrative example, AWS cites a portfolio of 250 vendor contracts of 10-20 pages each (up to 5000 pages in total), where manual processing of recurring management queries took over a week of an analyst's work. According to AWS, the described pipeline can process a single contract in a matter of seconds under typical conditions, and the serverless architecture is designed to handle parallel processing of many contracts at once.
The source article is only partially available, including details about the specific choice of the two models for extraction and verification. See the source article for further details.
Why it matters
Companies managing hundreds to thousands of contracts or other documents run into the same limitation AWS describes: common AI chat tools based on RAG give confident but incorrect answers to questions requiring aggregation across an entire dataset. The described architecture shows a concrete alternative — combining structured data extraction into a database with RAG for queries about individual documents — which architects and developers can use as a pattern when designing their own AI agentic systems for analyzing large sets of documents.
Two audiences, two different impacts
What this means
For individuals
Developers and architects designing AI agentic systems for working with documents gain a concrete reference pattern for when it makes sense to replace pure RAG search with structured data extraction into a database.
For a business
Companies with a large portfolio of contracts or similar documents can use the described architecture (AI extraction into a database + analytics + RAG for individual documents) to significantly shorten work with searching and aggregating data across a portfolio that until now was manual and took weeks.
ProcessesCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.