Artificial Intelligence

IskayPet Unifies Corporate Knowledge with Amazon Bedrock

The leading Iberian pet care company leverages managed RAG infrastructure and AI to centralize internal documentation across multiple countries and languages.

Radwan AbdulhadiEditor and technology writer

Published: October 22, 2026 · Updated: October 22, 2026 · 3 min read

IskayPet Unifies Corporate Knowledge with Amazon Bedrock — Image: AWS Machine Learning Blog

Tackling a Decentralized Knowledge Challenge

IskayPet, a prominent pet care company operating across Spain, Portugal, and France, faced significant hurdles regarding internal information retrieval. Operational answers regarding return policies, veterinary protocols, and onboarding steps were scattered across hundreds of Microsoft SharePoint documents in four different languages and three countries. Searching these large repositories manually proved time-consuming and often yielded outdated or incorrect information as the business expanded. Further details are available from AWS Machine Learning Blog in the original source material.

To address this operational bottleneck, the company needed a natural-language assistant capable of working across multiple languages without maintaining separate systems, accurately routing queries without cross-contamination, scaling easily, citing sources transparently, and reusing existing SharePoint infrastructure. Traditional Retrieval-Augmented Generation (RAG) approaches typically require managing complex vector databases, embedding pipelines, and ingestion jobs internally, prompting IskayPet to seek a fully managed alternative. Further details are available from Artificial Intelligence in the original source material.

Leveraging Managed RAG Infrastructure

The company chose Amazon Bedrock Managed Knowledge Bases to handle data ingestion, multimodal parsing, vector storage, embedding generation, and retrieval tasks. This configuration allows IskayPet to pay solely for stored and retrieved data while bypassing the operational overhead of running custom RAG infrastructure.

Key features utilized in the deployment include a native SharePoint connector that ingests documents directly from current repositories without requiring data migration. Furthermore, advanced indexing settings enable the extraction of content from images embedded within documents such as PDFs, DOCX, and PPTX files, making figures and scanned procedures fully searchable.

Hybrid Search and Security Controls

The underlying architecture incorporates built-in hybrid search, combining semantic search with keyword searches based on the BM25 algorithm at no extra cost. This combination successfully recovers exact product codes and specific references that pure semantic search methodologies might otherwise miss.

To maintain strict internal data governance, the knowledge base enforces document-level access control during retrieval. By respecting the source system's existing permissions, queries return only passages from documents that the requesting user is already authorized to read, avoiding the need to build a separate authorization model from scratch.

Solution architecture — Image related to the report from AWS Machine Learning Blog · Source: AWS Machine Learning Blog

Development with Kiro and Claude Haiku 4.5

IskayPet developed the entire solution using Kiro, an agentic development environment running on Claude Haiku 4.5. The development team used the tool to build the store-facing frontend interface, design the AWS architecture, and provision necessary services including knowledge base configurations, data sources, and IAM roles.

The combination of an agentic workflow and a fast, cost-effective foundation model allowed a small engineering team to transition the project from initial concept to production rapidly while maintaining code and infrastructure consistency.

Production Workflow and Evaluation

When an employee submits a natural-language query via the chat interface, the application passes the user identity to the knowledge base to ensure proper permission enforcement. The system then executes a hybrid search, reranks the retrieved results based on relevance, and generates a grounded response complete with source citations.

Employees can evaluate the responses using a thumbs up or thumbs down feedback mechanism, which feeds directly into the evaluation set. IskayPet continues to measure solution quality incrementally using built-in validation tools to maintain high accuracy across all business units.