Source URL: https://blog.scottlogic.com/2025/04/02/inferesg-finding-the-right-architecture.html
Source: Scott Logic
Title: InferESG: Finding the Right Architecture for AI-Powered ESG Analysis
Feedly Summary: During the InferESG project, we made a pivotal decision to use an orchestrated workflow to conduct a deterministic process, parallel to the agentic system.
AI Summary and Description: Yes
Summary: The text outlines the development journey of the InferESG project’s architecture, which integrates a hybrid approach combining deterministic workflows for sustainability report analysis with a conversational, agentic structure for dynamic exploration. Key insights include the challenges faced in initial conversations versus structured tasks and the potential of this hybrid solution to adapt to evolving ESG standards.
Detailed Description:
The text presents a comprehensive analysis of the InferESG project, which focuses on improving the analysis of environmental, social, and governance (ESG) sustainability reports. The project initially adopted a conversational framework using an Intent Agent and Supervisor for processing queries. However, it identified limitations in this approach when analyzing full reports. To enhance functionality, a more deterministic architecture was developed, ensuring structured, consistent outputs while allowing for dynamic interactions.
Major points of the analysis include:
– **Project Motivation**: The need to analyze and compare sustainability reports for companies led to a redesign of the architecture.
– **Initial Architecture**:
– Utilized an Intent Agent to parse user inquiries into smaller, manageable prompts.
– Faced issues in accurately processing complex, multifaceted report analysis, leading to inconsistent results.
– **Evolution to Deterministic Approach**:
– Removed the Intent Agent and Supervisor to construct a clear report workflow that specifies tasks directly.
– Established a synchronous and asynchronous processing method to ensure a repeatable and reliable analysis of sustainability reports.
– **Integration of Both Approaches**:
– Maintains the conversational framework for user engagement while using structured processes for comprehensive analysis.
– Allows analysts to explore specific aspects of reports dynamically, fostering deeper inquiries beyond the initial analysis.
Benefits of the hybrid approach include:
– **Consistency**: Ensures thorough coverage of essential ESG topics through structured methodology.
– **Flexibility**: The agentic component allows dynamic queries and responses catering to analyst inquiries.
– **Transparency**: Clear workflow stages enhance understanding and validation of the analysis process, important for compliance and governance.
– **Adaptability**: The system can evolve with changing ESG standards, providing continuous improvement opportunities.
In conclusion, the text emphasizes that blending deterministic workflows with dynamic agentic capabilities offers a robust solution for ESG report analysis. This architectural synergy is posited to be foundational for addressing future challenges in sustainability evaluation and reporting.