Multi-agent AI platform
Every output that Trubient produces can be traced to its source document, and the reasoning behind it can be explained. This transparency is essential for regulated investment decisions, which must be supported by verifiable evidence.
Every output that Trubient produces follows this principle. The platform provides 100% source attribution, which means that every data point is linked to the document and page from which it was taken.
Every data point links directly to its source document and page number. The user can follow the link to see the original text highlighted in the document.
When data is not available, the system states what it searched for and why the information could not be found. It does not estimate or invent a value to fill the gap.
Every extraction is timestamped and logged. The reasoning behind any result can be replayed at any time, which supports regulatory audits and internal review.
A single model that performs every task tends to perform each of them less well. Trubient uses specialised agents, each optimised for one task and coordinated by a supervisor agent.
Orchestration
Supervisor Agent
Document processing
Document Processor
Data extraction
Data Extractor
Workflow monitoring
Notification Agent
Orchestration
The Supervisor Agent orchestrates the workflow and routes each task to the appropriate specialist agent. It handles exceptions, maintains quality control across the entire pipeline and determines when an item should be escalated for human review.
Document processing
The Document Processor performs PDF parsing, optical character recognition (OCR), table extraction and document structure analysis. It is tuned for layout analysis and converts unstructured files into a machine-readable format.
Data extraction
The Data Extractor uses retrieval-augmented generation (RAG) to locate specific metrics with an understanding of their context, and it cross-references related data points. It is trained on financial terminology and sustainability frameworks.
Workflow monitoring
The Notification Agent monitors the status of the workflow and issues alerts when a task is completed or an exception occurs. It routes items that require human review to the appropriate members of the team.
The reason for a multi-agent architecture. A general-purpose language model can perform many tasks to an adequate standard, but the extraction of financial data for investment decisions requires a higher standard. Each Trubient agent is therefore optimised for its own task, and the Supervisor Agent coordinates their work. This specialisation provides higher accuracy and better explainability than a single model, and it allows unusual documents and other exceptional cases to be handled reliably.
The proprietary Trubient framework covers the environmental, social and governance pillars and is aligned with recognised international standards. The client's own methodology and weightings are applied to the extracted data.
134 indicators
Categories
102 indicators
Categories
85 indicators
Categories
Trubient is built on proven technologies and runs on Google Cloud. It is deployed through fully automated pipelines.
The large language models are trained and fine-tuned for financial documents.
A multi-agent system orchestrates the specialist agents that process documents and extract data.
The platform is built on Google Cloud by IQZ Systems, a Google Cloud partner.
The platform is based on open-source components, so the technology stack can be audited.
In a technical demonstration, the Trubient agents will process a document that you provide, and our team will answer your questions about how the platform works.
