The organisation stored a large volume of internal documents—policy files, reports, SOPs, manuals, contracts, and reference materials—in an Amazon S3 bucket. Employees frequently needed answers, summaries, and insights from these documents, but the process was slow and inefficient.
Key challenges included:
- Difficulty locating relevant information across hundreds of unstructured documents
- Time-consuming manual reading for summaries, insights, or decision-making
- Lack of a unified system that could provide simple answers, executive summaries, analytics, or suggestions based on the document contents
- Dependence on keyword search, which often returned irrelevant or incomplete results
- No AI interface for conversational access to internal knowledge
The organisation needed a secure, intelligent, conversational AI system that could understand document content and produce accurate, generative outputs on demand.
We built a Custom GPT solution powered by Amazon Bedrock Knowledge Base, seamlessly integrated with Open Web UI, enabling users to interact with internal documents using natural language.
The solution provides:
- Simple, accurate, context-aware answers sourced from internal documents
- Summaries of long files or sections
- Analytics such as key points, trends, risks, comparisons, and decisions
- Suggestion generation for improvements, next steps, or recommendations based on document content
- Automatic ingestion of documents uploaded to S3
- A user-friendly conversational UI for employees across departments
This transformed the document repository into an intelligent, generative, business-ready knowledge system.
A. Creating the Bedrock Knowledge Base
- Configured a Bedrock Knowledge Base linked to an S3 bucket
- Automated ingestion of all PDF, DOCX, TXT, and other supported file formats
- Enabled semantic chunking and embedding generation for precise context retrieval
- Implemented automatic, event-driven updates whenever new documents were uploaded to S3
B. Document Processing Pipeline
Preprocessed documents for:
- Text extraction
- Chunk division
- Metadata tagging
Created vector embeddings based on Bedrock’s built-in embedding models
Indexed chunks for fast semantic search and retrieval
C. Integrating Custom GPT Model
Connected the Knowledge Base with a Bedrock foundation model (Titan, Claude, or Llama)
Implemented Retrieval-Augmented Generation (RAG) so GPT always uses verified document context
Trained the system prompts to generate:
- Direct answers
- Concise or detailed summaries
- Analytical breakdowns (strengths, risks, compliance points, insights)
- Suggestions or recommendations grounded in document content
D. Open Web UI Integration
Connected backend APIs to a user-friendly Open Web UI
Delivered features such as:
- Chat-based interactions
- Content-based citations
- Document-specific responses
- Exportable summaries or answers
- Provided secure access management for authorized users
E. Security & Governance
- Applied IAM roles for secure S3 and Bedrock access
- Encrypted data in transit and at rest
- Enabled detailed logging for audit trails
- Added response guardrails to reduce hallucinations and ensure context-based accuracy
Productivity Boost
- Reduced document search and review time by 80–90%
- Users received instant answers, summaries, and insights instead of manually reading multiple files
Better Decision-Making
- Teams gained immediate access to analytics, key points, and recommendations
- Complex reports became easy to understand through structured summaries
Improved Knowledge Access
Employees could ask natural questions like:
- “Summarize the latest policy update.”
- “What are the risks mentioned in this document?”
- “Give me suggestions for improving this process.
No need for manual keyword search or document hunting
Scalability & Automation
- New documents uploaded to S3 were automatically indexed
- System scaled effortlessly as document volume grew
- No manual configuration required for new content
Consistency & Accuracy
- Responses grounded in actual document content improved compliance and reduced misinformation
- Provided reliable, repeatable answers aligned with internal standards
The Custom GPT solution built with Amazon Bedrock Knowledge Base, S3 integration, and Open Web UI successfully transformed the organization’s unstructured document repository into a powerful, generative knowledge engine. Users can now obtain simple answers, summaries, analytics, and suggestions instantly—directly derived from internal content.
This solution enhanced productivity, accelerated decision-making, and created a scalable, intelligent foundation for future AI innovations within the organization.
