Knowledge Access Transformation with RAG and AWS

Challenge

Every proposal and contract started with a search. Contract documents sat in SharePoint folders that each team organized its own way, drafts lived in S3, and deal history lived in Salesforce. Finding the right document meant knowing the client’s name and the folder, then working out which version was current, or asking the colleague who knew.

This led to:

• time spent searching several places for one answer
• contract data copied from outdated versions
• slower turnaround on proposals and contracts

The client needed one place to ask, with answers people could check against the source.

Solution We Delivered

Crunch-IS developed a Retrieval-Augmented Generation (RAG) system designed to unify and simplify enterprise knowledge access.

At its core, the solution integrates AWS Kendra as the primary semantic search engine, enriched with LLM capabilities through AWS Bedrock for context-aware querying and summarization.

Data from SharePoint, Salesforce, and S3 is continuously synchronized through AWS Step Functions, ensuring all sources remain up to date and searchable.

How an answer looks. A user types a question such as “What are our standard payment terms for enterprise contracts?” The assistant replies with a short answer, then lists the documents it used, each with a link to the original file. Filters narrow the results by customer, document type, or source. We built the interface with React.js and Nest.js.

Who sees what. The assistant answers only from documents the user is allowed to see. If you ask about a file you are not cleared to open, it replies that it has no such information.

Part of a wider enterprise AI assistant. The knowledge base is one tool in the client’s enterprise AI assistant, which gives sales, marketing, HR, legal, and project teams their own AI tools without sending company documents to public AI services. The same knowledge base runs MSA and compliance checks on a customer. The assistant also generates RFP documents from the company’s own templates and reviews presentations against house rules.

Implementation Approach

To ensure performance, scalability, and security, we deployed the solution on a cloud-native AWS architecture:

• AWS ECS – handled containerized deployment of microservices for efficient scaling.
• AWS Step Functions – orchestrated automated ETL workflows for continuous data ingestion.
• AWS RDS (PostgreSQL) – ensured reliable data persistence and structured metadata management.
• AWS Amplify – streamlined CI/CD and deployment for the user-facing app.

The architecture keeps the index in step with each source system, applies each user’s access rights to every answer, and can take new sources, such as Google Drive or uploaded files, through the same pipeline.

Client’s Results

Sales and delivery teams now start with a question, not a search:

• 60% faster work with company documents – sales and delivery teams find proposals, contracts, and project records in one search instead of hunting across systems.
• One place to ask – a single knowledge base replaced documents scattered across systems and team folders.
• Answers people can check – every answer links to its sources, so a fact can be verified before it goes into a contract.
• Fewer interruptions – people find a document themselves instead of asking the colleague who knows where it is.
• Faster onboarding – new hires ask the assistant instead of their colleagues.