Smart Widget for ESG Risk Assessment

Introduction

Expanded existing AI agent solution to provide more information regarding reasoning steps and enhance user experience while ensuring evaluation can be executed on sight.
The AI agent was enhanced to surface intermediate reasoning steps to users during multi-step queries. This improvement addressed transparency gaps and allowed both users and internal teams to follow the logic behind each action. With the introduction of real-time Server-Sent Events (SSE), the system transformed passive wait time into interactive engagement.

Problem

Their users often submitted complex queries that involved fetching and aggregating data across several services. These multi-step workflows could take time, but users were left without any indication of what was happening. This lack of visibility led to frustration and uncertainty. Moreover, poorly phrased queries would often yield irrelevant or vague responses, with no mechanism to guide users toward refinement.

Specific Pain Points:

• Long execution times for complex queries
• Users frustrated by no visible system progress
• Inability to distinguish between missing data and service failures
• No real-time insight into agent reasoning, only backend logs available
• Users are unaware when their query was ambiguous and needed clarification

Solution

To address these challenges, the team implemented Server-Sent Events (SSE) that emitted live status updates from each component of the agent’s process. These updates were displayed in a clean, user-friendly UI that illustrated the flow of reasoning step-by-step. This allowed users to see what the system was doing in real time and understand why a query might be taking longer to resolve or failing altogether.

LangChain and LangGraph powered the orchestration behind the scenes, allowing for modular, traceable workflows across tool invocations.

Specific Steps Taken:

• Designed a standardized payload format for SSE to unify status reporting across tools
• Built an intuitive frontend component to render live reasoning steps
• Integrated real-time feedback into each step of the agent’s processing
• Scaled the solution to support high user concurrency with reliable performance
• Ran end-to-end tests and user trials to validate functionality and usability

Unique Value Proposition:

• Working on a system in production with real users on a new technology in the quickly developing AI space.
• Users gain visibility into what the AI is doing at each step, improving trust and clarity.
• Clarification prompts during reasoning help users refine inputs on the fly.
• Internal teams can monitor and debug live agent workflows without engineering support.
• Enhanced transparency improves both user experience and internal evaluation workflows.

Results

Key Metrics:

• 30 – 50% reduction in user abandonment during long-running queries.
• Fewer support tickets about “unresponsive” bot behavior.
• More refined user queries with fewer repetitions.
• Faster developer response times due to improved visibility.

Smart Tip

Expose what your agent is doing, not just the final result. Showing progress builds trust and turns passive waiting into engaged exploration.

Smart Fact

Their real-time UI streams tool-level reasoning via Server-Sent Events, providing an “X-ray view” of what the agent is doing, previously only visible in backend logs.

About the Clients

This company is a global sustainability technology platform delivering ESG and sustainability insights to investors, companies, and consumers. Their mission is to bring societal impact to markets by enabling data-driven decision-making around environmental and social outcomes.

Technologies Used

• LangChain
• LangGraph
• Python (FastAPI)
• Server-Sent Events (SSE)
• Custom UI Components (Vue.js)
• Large Language Models (LLMs)