Electric Lion UI

About the Project

Inspiration

The inspiration for this project came from the challenges I faced while trying to stay on top of stock market news. With thousands of articles published daily across multiple sources, it's increasingly difficult for investors and analysts to efficiently find relevant information and understand its market implications. I wanted to create an AI-powered solution that could not only search through news effectively but also provide intelligent analysis and insights.

What it Does

This project combines modern AI technologies to create an intelligent stock market news assistant that can:

Search and retrieve relevant news articles across multiple sources
Analyze news content and provide market implications
Answer questions about specific stocks or market trends
Maintain context and provide source citations
Generate follow-up questions for deeper analysis

The system uses vector databases to enable semantic search capabilities, meaning it can understand the intent behind user queries rather than just matching keywords.

How I Built It

The project is built using several key technologies:

Vector Databases: Implemented using PostgreSQL with pgvector extension for efficient similarity search
Embedding Models: Utilized Together AI's embedding models for converting text into vector representations
LLM Integration: Integrated Google's Gemini model for natural language understanding and generation
Reranking: Implemented Cohere's reranking system to improve search result relevance
Document Processing: Created a robust system for processing and storing news articles with metadata

The architecture follows a modular design, allowing for easy updates and extensions. Each component (knowledge base, agent, embedder, etc.) is designed to be interchangeable and independently testable.

Challenges

Several significant challenges were encountered during development:

Data Management: Handling large volumes of news data efficiently while maintaining search performance required careful optimization of the vector database schema and indexing.
Context Management: Ensuring the AI maintains relevant context across conversations while not getting overwhelmed with information was a delicate balance.
Search Accuracy: Fine-tuning the search and reranking mechanisms to provide truly relevant results took significant iteration and testing.
Error Handling: Building robust error handling for various API calls and database operations while maintaining a smooth user experience required careful consideration.