Apppublic

crazyafro/financial_intelligence_agent

sourceHugging Facemitupdated 4mo agoView on Hugging Face

0likes

App README

Financial Intelligence Agent

A multi-agent AI platform for financial research, live market data, quantitative analysis, and automated report generation — built on LangGraph's supervisor pattern with a Gradio web UI.

[image]

Architecture

The system uses a Supervisor → Specialized Agent pattern. A central supervisor LLM reads every user message and decides which specialist agent to invoke. Agents call their tools, return results to the supervisor, and the loop continues until the task is complete.

User (Gradio UI)
      │
      ▼
 Supervisor  (GPT-4o-mini · routes tasks)
  ┌───┼───────┬──────────┬────────┐
  ▼   ▼       ▼          ▼        ▼
Research  Finance  Analysis  Report  RAG
 Agent    Agent    Agent     Agent   Agent
  │         │        │         │       │
Tools     Tools    Tools   Format  index_doc
                           ──↓──   search_kb
                         [HITL]
                           ──↓──
                          Save

Agents

Agent	Role	Tools
Supervisor	Routes requests to the best-fit agent; signals `FINISH` when done	—
Research Agent	Finds news, background info, and recent events	`web_search`, `wikipedia_search`
Finance Agent	Retrieves live market data	`get_stock_price`, `get_crypto_price`, `get_exchange_rate`
Analysis Agent	Performs calculations, date context, text summarization	`calculator`, `get_date_context`, `summarize_text`
Report Agent	Formats reports then waits for human approval before saving	`format_markdown_report`, `save_report_to_file`
RAG Agent	Searches and indexes the persistent vector knowledge base	`search_knowledge_base`, `index_document`

Tools

Tool	Description	Data Source
`web_search`	Real-time internet search (4 results)	Tavily API
`wikipedia_search`	Wikipedia summaries with disambiguation handling	Wikipedia
`get_stock_price`	Price, market cap, P/E ratio, 52-week range	yfinance
`get_crypto_price`	Price, market cap, 24h change	CoinGecko API
`get_exchange_rate`	Currency pair conversion rates	ExchangeRate-API
`calculator`	Safe math evaluation (`sqrt`, `log`, `pow`, etc.)	Python `math`
`get_date_context`	Current date, time, day, and fiscal quarter	System clock
`summarize_text`	Condenses text into a 3-5 sentence summary	GPT-4o-mini
`format_markdown_report`	Structures raw data into a professional Markdown report	GPT-4o-mini
`save_report_to_file`	Saves report to `reports/` — requires human approval first	Local filesystem
`search_knowledge_base`	Semantic search over past analyses and reports	ChromaDB + OpenAI Embeddings
`index_document`	Indexes text into the vector store for future retrieval	ChromaDB + OpenAI Embeddings

Features

RAG — Persistent Vector Knowledge Base

The RAG Agent stores and retrieves information from a local ChromaDB vector database (chroma_db/). Every saved report or key finding can be indexed, enabling the system to:

—Recall past analyses across sessions
—Avoid duplicate research
—Find related context from previous work

Human-in-the-Loop Report Approval

Before any report is written to disk, the graph pauses (via LangGraph's interrupt_before) and surfaces an approval panel in the UI. You can:

—Approve & Save — resumes the graph, writes the file to reports/
—Cancel Saving — injects a rejection message, the graph completes without saving

Setup

Prerequisites

—Python 3.9+
—OpenAI API key
—Tavily API key
—ExchangeRate-API key
—(Optional) LangSmith API key for tracing

Installation

bash

git clone <repo-url>
cd financial_intelligence_agent
pip install -r requirements.txt

Configuration

Create a .env file in the project root:

env

OPENAI_API_KEY=sk-...
TAVILY_API_KEY=tvly-...
EXCHANGE_RATE_API_KEY=...

# Optional: enables LangSmith tracing
LANGCHAIN_API_KEY=lsv2_...
LANGCHAIN_TRACING_V2=true
LANGCHAIN_PROJECT=financial-intelligence-agent

For Hugging Face Spaces, set these as repository secrets (Settings → Variables and secrets) instead of using a .env file.

Usage

bash

python app.py

Open the Gradio UI at http://localhost:7860.

Example Queries

Query	Agents Typically Used
Research Tesla, get stock price, convert market cap to EUR, save a report	Research + Finance + Analysis + Report
What is Bitcoin's price? Summarize recent news and save a report	Finance + Research + Analysis + Report
Compare Apple vs Microsoft stocks — which has a higher P/E?	Finance + Analysis
Search for the latest AI regulation news and format a briefing	Research + Report
What is 15% of Tesla's current stock price?	Finance + Analysis
Search the knowledge base for past analyses on Apple	RAG

UI Controls

Control	Description
Show Agent Reasoning Trace	Expands all tool calls and raw results below the answer
New Session	Starts a fresh conversation with a new thread ID
Session Thread ID	Keep this to resume a conversation with full memory
Approve & Save	(appears when a report is ready) Confirms writing to disk
Cancel Saving	(appears when a report is ready) Discards the save without losing the formatted content

Memory and Persistence

Store	Technology	Location	Persists across restarts
Conversation memory	SQLite via LangGraph `SqliteSaver`	`agent_memory.db`	Yes
Knowledge base (RAG)	ChromaDB + OpenAI Embeddings	`chroma_db/`	Yes
Generated reports	Markdown files	`reports/`	Yes

Note for Hugging Face Spaces: the Spaces filesystem is ephemeral and resets on container restart. For persistent storage on Spaces, mount a Persistent Storage volume or use an external database.

Project Structure

financial_intelligence_agent/
├── app.py                    # Gradio web interface — entry point
├── requirements.txt
├── .env                      # API keys (not committed — see .gitignore)
├── .gitignore
├── diagram.png               # Architecture diagram
├── generate_diagram.py       # Script to regenerate diagram.png
│
├── agents/
│   ├── __init__.py           # Exports all agent nodes
│   ├── state.py              # Shared AgentState TypedDict
│   ├── supervisor.py         # Supervisor routing logic
│   ├── research_agent.py     # Web search + Wikipedia
│   ├── finance_agent.py      # Stock, crypto, FX data
│   ├── analysis_agent.py     # Calculator, date, summarize
│   ├── report_agent.py       # Format + save reports (HITL gated)
│   └── rag_agent.py          # Vector store search + index
│
├── graph/
│   ├── __init__.py           # Exports build_graph, get_memory
│   ├── workflow.py           # LangGraph StateGraph assembly + interrupt
│   └── memory.py             # SQLite checkpointer
│
├── tools/
│   ├── __init__.py           # Aggregates all tools by category
│   ├── search_tools.py       # web_search, wikipedia_search
│   ├── finance_tools.py      # get_stock_price, get_crypto_price, get_exchange_rate
│   ├── analysis_tools.py     # calculator, get_date_context, summarize_text
│   ├── report_tools.py       # format_markdown_report, save_report_to_file
│   └── rag_tools.py          # index_document, search_knowledge_base
│
├── reports/                  # Auto-created; timestamped .md reports
├── chroma_db/                # Auto-created; ChromaDB vector store
└── agent_memory.db           # Auto-created; SQLite conversation memory

Deploying to Hugging Face Spaces

This repo is pre-configured for Spaces deployment (see YAML frontmatter above and server_name="0.0.0.0" in app.py).

Steps (requires you to do once):

1.Push this repo to a GitHub repository
2.Go to huggingface.co/new-space → choose Gradio SDK → link your GitHub repo
3.In Space Settings → Variables and secrets, add your API keys:
4.OPENAI_API_KEY
5.TAVILY_API_KEY
6.EXCHANGE_RATE_API_KEY
7.(Optional) Add a Persistent Storage volume so chroma_db/, reports/, and agent_memory.db survive container restarts

Tech Stack

Component	Technology
LLM	OpenAI GPT-4o-mini
Agent framework	LangChain + LangGraph
Vector store (RAG)	ChromaDB + OpenAI Embeddings (`text-embedding-3-small`)
Web UI	Gradio
Conversation memory	LangGraph SQLite checkpointer
Web search	Tavily
Stock data	yfinance
Crypto data	CoinGecko API
Currency data	ExchangeRate-API
Tracing	LangSmith (optional)