The AI Agentic Financial Assistant built with Model Context Protocol

Overview

ASSYST partnered with a Global Development Financial Institution to develop an advanced AI-based Budget and Business Intelligence Chatbot. Powered by the Microsoft AI Stack, this solution is designed to streamline financial management, risk assessment, and budgeting processes across corporate and global development projects. The chatbot integrates a Chat UI into users' daily routines and workflows, enhancing employee productivity by automating routine tasks and providing secure, role-based access to information. It also centralizes data, supports self-service capabilities, and ensures interoperability with existing systems. The chatbot provides real-time insights and scalable solutions by enhancing knowledge management and facilitating effective process discovery, thereby significantly improving operational efficiency and decision-making throughout the institution.

Challenges

The institution faced several significant challenges. Business administration, IT, and financial data were dispersed across multiple systems, making it difficult to catalog and present a unified view. Traditional manual methods of aggregating reports were time-consuming, error-prone, and resource-intensive, hampering operational efficiency. As the institution expanded and diversified, managing and locating data effectively became increasingly complex.

Solution

ASSYST proposed an intuitive and intelligent chatbot application to address these challenges. The chatbot automates the discovery of FAQs by continuously analyzing user queries and system interactions, providing information about systems by accessing relevant databases and documents, and extracting insights from diverse data sources using advanced AI algorithms. This approach reduces the time and effort required for analysis, thereby improving overall operational efficiency.

Key features of the solution include:

• Streamlined Automation: The chatbot utilizes Azure AI Search to extract insights and identify frequently asked questions, thereby significantly enhancing operational efficiency.
• Specialized AI Assistants: Multiple bots were developed, each tailored to specific focus areas. These bots facilitate human-like conversations with enterprise data, including documents, making it easier for users to interact with and extract information from complex datasets.
• LLM Capabilities: By integrating Large Language Models (LLMs) such as Azure OpenAI’s GPT-4, the chatbot can summarize information, answer basic queries, and perform semantic searches, improving accessibility to insights.
• The chatbot’s interface is designed with the user in mind, ensuring that technical knowledge is not a barrier to accessing data and insights. This user-friendly design makes the system easily accessible to all users, promoting a sense of comfort and ease.
• Agentic AI with Model Context Protocol (MCP): To further elevate intelligence and adaptability, the chatbot incorporates Agentic AI powered by the Model Context Protocol (MCP). This structured framework allows the AI to autonomously retrain, update, and apply relevant contextual knowledge across interactions. As a result, the chatbot delivers smarter, more personalized responses, thereby minimizing redundant processing, improving performance & deepening its understanding of user intent.

Agentic AI Architecture

The solution is implemented across three distinct channels:

1. Chatbot for Analytics: This application enables users to gain insights into complex data through intelligent, conversational interactions. Integration with Microsoft Teams enhances accessibility and usability.
2. FAQ Bot: Designed for corporate budgeting needs, this bot provides intelligent responses to frequently asked questions related to procurement, budgeting, and administration. Using the Azure AI-Language service, it retrieves accurate information relevant to user queries.
3. Enterprise Application Bot: Integrated with various enterprise applications, this bot offers brief descriptions and links to relevant applications, guiding users based on their specific needs.

Notable features include Microsoft Teams integration and functionalities similar to ChatGPT, such as PDF summarization, data extraction from PDFs, and database connectivity. Users can provide feedback, and the system suggests actions such as initiating a call to the IT team, directing users to IT Helpdesk chat support, or offering email support.

Technology Stack

The project utilized an advanced technology stack to support the solution. The core AI infrastructure was based on the Microsoft Azure Stack, with LLM capabilities provided by Azure OpenAI (GPT-4). The user interface was built using ReactJS, ensuring an intuitive and responsive experience. The Semantic Kernel handled AI orchestration and meta-prompt management, while Azure Search, enhanced with Cognitive Search, provided memory and search capabilities. Throughout the project, data security and trust were given top priority. Institutional data remained secure, with fine-tuned data securely maintained within the subscription, emphasizing data sovereignty and customer copyright for applications built on the Azure OpenAI Service. This approach underscores our commitment to implementing ethical and responsible AI, ensuring the system's safety and reliability.

Impact

The AI-based Budget and Business Intelligence Chatbot has profoundly transformed the user experience of financial and budgeting systems by introducing innovative interaction modes. This solution has enhanced financial insights, streamlined budgeting processes, and improved risk management by addressing challenges such as scattered systems, manual inefficiencies, and increasing complexity. Beyond these improvements, the chatbot has significantly elevated broader knowledge gain and process literacy within the organization. By offering intuitive access to complex data and delivering clear, actionable insights, it has empowered users to better understand and navigate financial systems, resulting in a more informed and capable workforce. The chatbot’s capabilities in autonomous data analysis, dynamic reporting, and user-friendly interaction have driven sustainable improvements and increased overall efficiency.

ASSYST AI Services

The ASSYST AI Center of Excellence (CoE) partners with customers and their internal business and technology teams to embed AI expertise, enhancing agility and growth as AI matures. The CoE provides AI resources, training, and best practices, embedding data scientists, machine learning experts, and product managers into project teams to work closely with domain experts. Clear communication channels ensure ongoing collaboration and knowledge sharing.

The CoE evaluates Agentic AI use cases, prioritizing them based on impact, feasibility, and strategic alignment. These use cases may include automating repetitive tasks, enhancing customer experiences, improving communication strategies by identifying user behavior patterns, and detecting fraud or anomalies in financial transactions. For instance, AI can be used to automate the identification and flagging of potential security threats, thereby enhancing the efficiency of Security Operations Center (SOC) teams. 

We leverage the ASSYST Green Accelerator Program and solutions like Collab AI and ComplySyncAI to deliver accelerators that support these use cases. We focus on data collection, preprocessing, and ensuring the quality and privacy of the data. Model development leverages algorithms such as decision trees, support vector machines, and neural networks with precise training, validation, and performance assessment. By incorporating and committing to human-centered design principles, we deliver intuitive, user-friendly, responsible, and humane AI solutions that meet the needs of end-users.

Let’s discuss your Agentic AI Use Case.