🪐 Kepler Financial Analyst
An advanced multi-agent AI system with a hybrid RAG pipeline for deep analysis of SEC filings.
Project Overview
Kepler is an advanced web application that answers complex questions about corporate financial documents. The system uses a hybrid RAG architecture, enriching semantic search with a dynamically constructed Knowledge Graph. This allows the AI to understand not just the text, but the relationships between key financial entities, leading to highly accurate and context-aware responses, including on-the-fly data visualizations.
Hybrid GraphRAG
Deep Contextual Insight
Key Features
🤖 Multi-Agent System
Uses specialist AI agents for routing, data extraction, and text generation for higher accuracy.
🕸️ Knowledge Graph
Dynamically builds a knowledge graph from the text to understand relationships between entities.
📊 On-the-Fly Visualizations
Translates natural language questions into interactive plots using Plotly.
🎯 Quantitative Evaluation
Achieved an 80% Factual Correctness Score using a rigorous, LLM-as-judge (RAGAS) evaluation.
Interactive Hybrid RAG Architecture
Click on each component of the pipeline to learn more about its role.
Document Processing
Text & Table Extraction
Vector Database
Semantic Search
Knowledge Graph
Relational Context
Multi-Agent System
Router & Specialists
Final Output
Answer or Chart
Select a component to see details.
Skills Demonstrated
- ✓Multi-Agent AI Systems
- ✓Hybrid RAG (Vector + Graph)
- ✓Knowledge Graph Construction
- ✓LLM-based Evaluation (RAGAS)
- ✓LLM Orchestration (LangChain)
Technology Stack
Quantitative Evaluation
on a blind test set
LLM-as-Judge Methodology
To ensure the system's reliability, a rigorous evaluation was performed. A "golden dataset" of question-answer pairs was created manually. The system's generated answers were then compared against this ground truth using a powerful LLM as an impartial judge. This RAGAS-style approach provides a quantitative measure of the system's factual accuracy, moving beyond subjective evaluation.