iHub RAG

Customer context

The partner serves US-based SMB and Enterprise clients across sectors such as Healthcare & Life Sciences, Financial Services, and Retail, operating teams across North and South America, the UK, and South Africa, while working US hours to match customer needs, and offering services spanning product development, cloud infrastructure, integrations/automation, security, and AI/ML on AWS. The business objective was to modernize user enablement and support through an embedded assistant grounded in the company’s living documentation and technical wiki stored in S3 and orchestrated by Bedrock Knowledge Bases.

Business challenge

Documentation existed across wikis, PDFs, legacy guides, screenshots, and user guide videos, making manual search slow and inconsistent while driving higher support costs and risking SLA breaches at scale. Legacy and multimodal materials reduced self-service effectiveness and complicated ingestion, making it difficult to achieve complete coverage in early phases of the project.

Solution overview

Amazon Bedrock Knowledge Bases managed ingestion, embeddings, retrieval, reranking, and prompt augmentation over S3-sourced documentation, enabling context-grounded answers with citations and rapid updates when content changes.

Amazon Bedrock Agents provided agentic orchestration for multi-step resolution, including tool invocation and iterative retrieval, exposed as a chatbot inside the SaaS UI through the AWS SDK for seamless user interaction.

Amazon S3 served as the durable, scalable document repository for the wiki and technical assets, enabling versioning and continuous ingestion for evolving content.

Architecture highlights

The SaaS UI calls the Bedrock Agent via the AWS SDK; the agent queries the Knowledge Base, retrieves top-ranked chunks, augments the prompt, and returns an answer with source attributions for verification and compliance review.

Bedrock Agents handle task planning and invocation of retrieval tools, reducing custom orchestration code and centralizing operational controls in managed services for resiliency and observability.

Knowledge Base features like reranking and citations support trustworthy outputs and enable quick reflection of documentation updates without retooling the application layer.

Implementation details

The team used Bedrock Knowledge Bases to avoid building bespoke ingestion and retrieval pipelines, accelerating time-to-value and simplifying operations with managed vectorization and retrieval augmentation.

Retrieval parameters and indexing strategies were tuned iteratively to balance answer quality and latency, following field techniques to improve RAG performance on Bedrock KBs.

Continuous ingestion and versioning in S3 ensured new documents and edits are reflected rapidly, with change control for auditability and rollback when necessary.

Outcomes and KPIs

Response times dropped from manual minutes to seconds with grounded citations, reducing escalations and enhancing user trust in answers surfaced from authoritative sources.

Automated resolution of common queries increased, lowering support cost and enabling specialists to focus on complex cases, while managed RAG/agent capabilities reduced operational overhead for the engineering team.

The solution established a sustainable path to expand into multimodal content as stronger models become available, guided by a pilot-and-measure approach to maintain cost controls.

Challenges and mitigations

Multimodal accuracy: legacy screenshots and video guides reduced extraction quality; access to improved models was requested and will be tested through staged pilots to balance accuracy and cost before broad rollout.

Legacy formats: preprocessing, quality scoring, and phased ingestion were implemented, and documentation standards were updated to favor structured, text-first content for future sources while retaining visual assets as supplements.

Security and compliance

All data flows remain inside AWS services with IAM-based least-privilege controls and managed encryption, leveraging Bedrock’s agent and knowledge base permissions for consistent security and auditability.

Citations enable traceability from answers back to sources, supporting internal review and compliance workflows in regulated sectors like healthcare and finance where document provenance is essential.

Why AWS

Bedrock Knowledge Bases and Agents offer managed RAG and agentic orchestration, session handling, retrieval with reranking, and citations—reducing custom infrastructure and operational risk while accelerating delivery timelines.

Native integration with S3 and the AWS SDK streamlines ingestion, updates, and secure access patterns, enabling an enterprise-grade document assistant with explainability and continuous refresh on AWS.