Angel Borroy

Building an AI-Ready Content Lake: Scaling RAG and Document AI Beyond Demos

How do you build a secure, permission-aware RAG system for your private enterprise data?

Building an AI-Ready Content Lake: Scaling RAG and Document AI Beyond Demos
#1about 3 minutes

Overcoming the limits of traditional enterprise search

Traditional keyword search fails across multiple content repositories, creating a need for a unified, semantic, and permission-aware solution.

#2about 5 minutes

Introducing the Hyland AI Ready Hub for on-premise RAG

The Hyland AI Ready Hub is an open-source, on-premise platform for building a multi-source, permission-aware RAG system to ensure data sovereignty.

#3about 4 minutes

Understanding the system architecture and data flow

The hub sits between knowledge enrichment and discovery, using batch and live ingesters to pull data from sources like Alfresco and Nuxeo into a central RAG service.

#4about 3 minutes

How to build custom connectors for new content sources

Developers can integrate new repositories by implementing four key interfaces for document representation, API communication, text extraction, and content scoping.

#5about 5 minutes

Implementing a robust two-phase document ingestion pipeline

The idempotent ingestion process first synchronizes metadata and permissions, then uses a transformation queue for asynchronous text extraction and chunking.

#6about 4 minutes

Improving retrieval with asymmetric embeddings and hybrid search

Retrieval accuracy is improved by using asymmetric embedding models for queries versus documents and by combining semantic vector search with traditional BM25 keyword search.

#7about 2 minutes

Exploring the RAG pipeline and source citation logic

The RAG pipeline processes user queries through hybrid search to build a context-rich prompt for an LLM, providing streaming answers with citations that link back to the original documents.

#8about 6 minutes

Live demo of a cross-repository search and chat

A live demonstration showcases the unified search and chat interface querying both Alfresco and Nuxeo repositories simultaneously while respecting user permissions.

#9about 3 minutes

Reviewing the deployment stack and hardware needs

The entire system is deployed via a Docker Compose stack of over 20 services, with recommendations for GPU-accelerated hardware and options for CPU-only development.

#10about 4 minutes

Recapping key benefits and core technical patterns

The solution provides AI-powered search without data migration or permission model changes by using key patterns like storage-layer permissions and asymmetric embeddings.

#11about 2 minutes

Outlining the future roadmap and deployment models

The future roadmap includes support for full on-premise, hybrid, and full-cloud deployments to provide maximum flexibility for enterprise needs.

Related jobs
Jobs that call for the skills explored in this talk.

Featured Partners

Related Articles

View all articles

From learning to earning

Jobs that call for the skills explored in this talk.