The rise of open-source AI tools has made it easier than ever to build sophisticated, cost-effective solutions. In this guide, I’ll show you how to create an AI-driven stock analysis system leveraging powerful open-source tools like LangGraph, DeepSeek R1–7B, and Ollama — all hosted on Google Colab to minimize infrastructure costs.

Whether you’re a data scientist, software developer, or just someone interested in AI, this workflow is designed to be modular, scalable, and developer-friendly.

The Case for Open Source AI

The democratization of large language models (LLMs) has opened exciting possibilities for developers:

• Free Compute Resources: Use platforms like Google Colab.
• Cost-effective Models: Avoid hefty cloud API costs by running models locally with Ollama.
• Customizability: Full control over models and pipelines.
• Learning Potential: Experiment hands-on with tools like DeepSeek.

By combining these tools, you can create systems that rival enterprise-grade setups at a fraction of the cost.

Why Run LLMs Locally?

Ollama allows you to run DeepSeek R1–7B locally, even in a Colab environment, with several advantages:

1. Cost-Effective: No reliance on costly API subscriptions.
2. Data Privacy: Sensitive stock data stays local.
3. Flexibility: Customize workflows and models to suit your needs.
4. Scalability: Expand functionality without vendor lock-in.

System Overview

Our AI pipeline follows a four-step architecture to deliver insights about any publicly traded stock.

Workflow:

1. Technical Analysis: Evaluates stock price patterns and market indicators.
2. Market Analysis: Reviews sector performance and comparative metrics.
3. News Analysis: Processes sentiment and breaking news events.
4. Recommendation: Synthesizes insights into actionable advice.

Each stage operates as an independent “Agent” and is part of a Directed Acyclic Graph (DAG) workflow.

Tools We’ll Use

Here’s a breakdown of the core technologies:

• LangChain: Framework for chaining AI prompts and agents.
• LangGraph: For orchestrating workflows with state-sharing across agents.
• DeepSeek R1–7B: Open-source LLM for analytical reasoning.
• Ollama: Local LLM hosting for cost savings and data privacy.
• Google Colab: Free, flexible compute environment.

These tools combine the power of modern LLMs with practical, developer-friendly workflows.

Implementing the Stock Analysis Pipeline

The heart of this system is the Directed Acyclic Graphs(DAG) workflow powered by LangGraph. Each agent in the workflow focuses on a specific analysis area.

1: Define the Workflow

def create_analysis_graph() -> Graph:
    """Create the analysis workflow graph"""
    # Create workflow graph
    workflow = StateGraph(State)

    # Add nodes
    workflow.add_node("technical", technical_analysis)
    workflow.add_node("market", market_analysis)
    workflow.add_node("news", news_analysis)
    workflow.add_node("recommendation", generate_recommendation)    # Define edges
    workflow.add_edge("technical", "market")
    workflow.add_edge("market", "news")
    workflow.add_edge("news", "recommendation")
    workflow.add_edge(START, "technical")    # Set end node
    workflow.add_edge("recommendation", END)

2: Build Each Agent

Here’s how to define a sample agent, like the Technical Analysis Agent:

# Technical Analysis Node
def technical_analysis(state: State) -> State:
    """Node for technical analysis"""
    symbol = state["symbol"]
    llm = state["llm"]

    # Fetch technical data
    stock = yf.Ticker(symbol)
    hist = stock.history(period='1y')    # Calculate indicators
    sma_20 = hist['Close'].rolling(window=20).mean()
    sma_50 = hist['Close'].rolling(window=50).mean()
    rsi = calculate_rsi(hist['Close'])    data = {
        'current_price': hist['Close'].iloc[-1],
        'sma_20': sma_20.iloc[-1],
        'sma_50': sma_50.iloc[-1],
        'rsi': rsi.iloc[-1],
        'volume_trend': hist['Volume'].iloc[-5:].mean() / hist['Volume'].iloc[-20:].mean()
    }    prompt = PromptTemplate.from_template(
        """Analyze these technical indicators for {symbol}:
        {data}        Provide:
        1. Trend analysis
        2. Support/Resistance levels
        3. Technical rating (Bullish/Neutral/Bearish)
        4. Key signals
        """
    )    chain = LLMChain(llm=llm, prompt=prompt)
    analysis = chain.run(symbol=symbol, data=json.dumps(data, indent=2))    state["results"]["technical"] = {
        "data": data,
        "analysis": analysis
    }
    return state

Similar steps for Market Analysis, News Analysis, and Recommendation Generation Agents.

Explore the Code

The repo includes:

• Full Colab notebook.
• Modular agent workflows for easy customization.

Real-World Applications

Once deployed, this pipeline provides comprehensive insights:

• Technical Indicators: Analyze moving averages, RSI, MACD, and more.
• Market Trends: Understand sector-wide or macroeconomic impacts.
• News Sentiment: Detect bullish or bearish trends in breaking news.
• Actionable Advice: Generate recommendations with confidence scores.

Final Thoughts

This guide shows how LangGraph, DeepSeek, and Ollama unlock new possibilities in AI-powered financial analysis. With open-source tools and free compute resources, you can build robust systems to tackle real-world challenges.

Important: AI-based recommendations are a supplement, not a substitute, for professional financial advice. Always conduct thorough research before making investment decisions.

Let’s build something incredible together! 🚀, and I look forward to connecting with fellow AI enthusiasts and professionals!