Skip to main content

​
Welcome to Finance Agent

Finance Agent is an AI-powered equity research platform that provides instant answers from authoritative financial documents. Ask questions and get precise, cited responses from 10-K filings, earnings calls, and market news.
Live Platform: www.stratalens.ai

​
What is Finance Agent?

Unlike generic LLMs that rely on web content, Finance Agent uses the same authoritative documents that professional analysts depend on:
  • Earnings Transcripts (2020-2025) - Word-for-word executive commentary from earnings calls
  • SEC 10-K Filings (2018-2025) - Official annual reports via specialized retrieval agent
  • Real-Time News - Latest market developments via Tavily search
  • Financial Screener - Natural language queries over company fundamentals (in development)

​
How it works

Finance Agent uses an advanced Retrieval-Augmented Generation (RAG) system with three key innovations:

Semantic routing

Routes to data sources based on question intent, not keywords. Questions about executive compensation automatically use 10-K filings, while quarterly performance questions search earnings transcripts.

Research planning

The agent explains its reasoning before searching: “I need to find Azure revenue figures, management commentary on competitive positioning, and forward guidance.”

Iterative improvement

Evaluates answer quality and searches for additional context until confident. Achieves 91% accuracy on the FinanceBench benchmark.

​
Core features

Multi-source RAG

Combines earnings transcripts, SEC 10-K filings, and news into unified responses with proper citations.

Specialized SEC agent

Dedicated retrieval agent for 10-K filings with section-level routing, table selection, and hybrid search. 91% accuracy on FinanceBench.

Parallel processing

Multi-ticker questions run per-company searches in parallel, then synthesize results into comparative analysis.

Real-time streaming

Watch the agent’s reasoning, see retrieved sources, and get answers progressively as they’re generated.

​
Performance

Benchmark Results:
  • 91% accuracy on FinanceBench (112 10-K questions)
  • Average response time: ~10 seconds per question
  • Evaluated using LLM-as-a-judge methodology

​
Example queries

Finance Agent handles diverse financial questions: 
"What was Apple's Q4 2024 revenue?"
→ Routes to earnings transcripts

"What is Tim Cook's compensation?"
→ Routes to 10-K filings (executive compensation only in SEC reports)

"Show me Microsoft's balance sheet"
→ Routes to 10-K financial statements

"Compare MSFT and GOOGL cloud revenue"
→ Parallel search across both companies with synthesis

"What's the latest news on NVIDIA?"
→ Routes to real-time news search

​
Architecture

The agent executes a 6-stage pipeline for each question:
1

Setup & initialization

Load RAG components, search engines, and available data quarters
2

Question analysis

Extract tickers, time periods, and semantic intent. Route to appropriate data sources (transcripts, 10-K, news, or hybrid).
3

Research planning

Generate reasoning statement and search plan. Resolve temporal references like “last 3 quarters” to specific dates.
4

Data retrieval

Execute hybrid vector + keyword search. For 10-K questions, invoke specialized SEC agent with iterative retrieval.
5

Iterative improvement

Evaluate answer quality (completeness, specificity, accuracy). Generate follow-up searches if needed. Stop when confidence threshold met.
6

Final assembly

Stream final answer with citations, source attributions, and metadata (confidence scores, chunks used).

​
Tech stack

  • Backend: FastAPI, PostgreSQL (pgvector), DuckDB
  • AI/ML: Cerebras (Qwen-3-235B), OpenAI (fallback), iterative self-improvement
  • Search: Hybrid vector (pgvector) + TF-IDF with cross-encoder reranking
  • Frontend: React + TypeScript, Tailwind CSS

​
Getting started

Quickstart

Get from zero to your first query in minutes

Installation

Complete installation guide with all dependencies

Agent system

Deep dive into RAG architecture and semantic routing

API reference

Explore REST and WebSocket endpoints

​
Use cases

Finance Agent is designed for:
  • Equity research analysts - Quick access to earnings call commentary and SEC filing data
  • Portfolio managers - Multi-company comparisons and trend analysis
  • Financial advisors - Client-ready summaries with proper citations
  • Developers - API access for building financial applications
  • Researchers - Benchmark dataset for financial QA systems

​
What’s next?

Explore the documentation to learn more:
Finance Agent is designed for research and analysis. Always verify critical information with original source documents before making investment decisions.

Build docs developers (and LLMs) love

Get started for freeTalk to us