Introduction
In the fast-paced world of financial markets, access to accurate, real-time information is crucial. Investing.com, a leading financial markets platform serving over 46 million monthly users across 44 language editions, recognized the need for an intelligent solution to help their subscribers navigate the complex landscape of financial data. This case study explores the development of an AI-powered financial assistant that would transform how users interact with financial information.
Problem Statement
Investing.com needed a sophisticated AI chatbot capable of:
- Interfacing with their internal financial query API (FinQL) to access metrics for over 70,000 financial entities
- Screening stocks based on various financial parameters
- Retrieving and synthesizing relevant financial news, analyst reports, and earnings call transcriptions
- Handling complex queries that require data from multiple sources
- Scaling to support thousands of concurrent users
The challenge was to create a versatile, user-friendly assistant that could understand and respond to a wide range of financial queries with accuracy and insight.
Challenges
Text-to-FinQL Translation
The FinQL API provides access to over 1,000 metrics for more than 70,000 financial entities. Translating natural language queries into precise API calls required a sophisticated approach beyond simple prompt engineering.
Financial Expertise in Prompt Engineering
While AI developers could handle basic prompt engineering, crafting high-quality responses demanded input from financial experts who could interpret complex financial data and provide meaningful insights.
Multi-Source Query Handling
Many user queries required information from multiple sources, necessitating a system that could dynamically aggregate data from financial metrics, news articles, and other relevant sources.
Scalability and Performance
The system needed to handle a high volume of concurrent users without compromising on speed or accuracy.
Solutions
Entity Extraction System
We developed an LLM-powered entity extraction system to identify relevant financial entities and metrics from user queries, enabling accurate data retrieval from the FinQL API.
Playground Gradio App
We created a Gradio-based application that allowed non-technical team members to test and refine chatbot responses. This tool provided full control over LLM selection, parameters, and system prompts, facilitating rapid iteration and improvement.
Dynamic RAG System
To address complex queries, we implemented a dynamic Retrieval-Augmented Generation (RAG) system. This solution analyzes user queries to select appropriate data sources, retrieves relevant information, and aggregates it for the LLM to generate comprehensive responses.
Scalable Cloud Infrastructure
We designed a robust, cloud-based infrastructure utilizing optimized database queries, caching mechanisms, load balancing, and auto-scaling features to ensure smooth performance under high demand.
Results
Rapid MVP Development
We delivered a functional Minimum Viable Product (MVP) in just two months, showcasing our ability to quickly transform concepts into working solutions.
Efficient Internal Testing and Iteration
Our in-house playground app empowered non-technical team members to refine the chatbot's responses, significantly enhancing the quality and accuracy of AI-generated financial insights.
Scalable Infrastructure
Successfully scaled the system to handle 1,000 concurrent users, ensuring robust performance and positioning the product for future growth.
Timely Product Launch
Soft-launched the product to InvestingPro users within nine months, allowing for real-world testing and user feedback collection.
Enhanced User Experience
The AI assistant significantly improved user engagement and satisfaction by providing quick, accurate responses to complex financial queries.
By leveraging cutting-edge AI technologies and financial expertise, this project delivered a powerful, scalable solution that revolutionized how Investing.com's users interact with financial data.