Data Analysis with AI

from langchain_anthropic import ChatAnthropic
from langchain_daytona_data_analysis import DaytonaDataAnalysisTool
from langchain.agents import AgentExecutor, create_tool_calling_agent
from langchain.prompts import ChatPromptTemplate
import os

# Initialize LangChain components
llm = ChatAnthropic(
    model="claude-3-5-sonnet-20241022",
    api_key=os.getenv("ANTHROPIC_API_KEY")
)

# Create Daytona data analysis tool
tool = DaytonaDataAnalysisTool(
    api_key=os.getenv("DAYTONA_API_KEY")
)

# Upload dataset with description
with open("dataset.csv", "rb") as f:
    tool.upload_file(
        file=f,
        description="""Vehicle valuations dataset with columns:
        - year: Manufacturing year
        - price_in_euro: Vehicle price
        - mileage_in_km: Kilometers driven
        
        Data cleaning needed:
        - Remove rows with missing values
        - Filter outliers in year and price
        """
    )

# Create agent with data analysis tool
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a data analyst. Use the data analysis tool to answer questions."),
    ("human", "{input}"),
    ("placeholder", "{agent_scratchpad}"),
])

agent = create_tool_calling_agent(llm, [tool], prompt)
agent_executor = AgentExecutor(agent=agent, tools=[tool], verbose=True)

# Run analysis with natural language
result = agent_executor.invoke({
    "input": """Analyze how vehicle prices vary by manufacturing year.
    Create a line chart showing average price per year."""
})

print(result["output"])

# Download generated charts
for artifact in tool.artifacts:
    if artifact.endswith('.png'):
        data = tool.download_file(artifact)
        with open(artifact, 'wb') as f:
            f.write(data)
        print(f"Chart saved to {artifact}")

# Clean up
tool.close()

Result stdout Original dataset shape: (100000, 15)
After removing missing values: (100000, 15)
After removing outliers: (90095, 15)

Average price by year:
year
2005.0     5968.12
2010.0    10378.82
2015.0    18563.86
2020.0    29400.16
2023.0    33119.84

Chart saved to chart-0.png

import openai
import os
from daytona_sdk import Daytona

client = openai.OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
CODING_MODEL = "gpt-5.1"
SUMMARY_MODEL = "gpt-4o"

# Generate analysis code
data_format = """CSV with columns: date, product_name, quantity, price"""
user_prompt = """Give the three highest revenue products for January 
                 and show them as a bar chart."""

code_response = client.chat.completions.create(
    model=CODING_MODEL,
    messages=[{
        "role": "user",
        "content": f"""Data format: {data_format}
        Task: {user_prompt}
        
        Generate Python code using pandas and matplotlib.
        Save charts as 'chart-N.png'."""
    }]
)

generated_code = code_response.choices[0].message.content

# Execute in Daytona sandbox
daytona = Daytona(api_key=os.getenv("DAYTONA_API_KEY"))
sandbox = daytona.create_sandbox()

# Upload dataset
sandbox.upload_file(
    local_path="cafe_sales_data.csv",
    remote_path="/home/daytona/data.csv"
)

# Install dependencies and run code
sandbox.exec("pip install pandas matplotlib")
result = sandbox.exec(f"python -c '{generated_code}'")

print(f"✓ Analysis complete")
print(result.stdout)

# Download charts
for file in sandbox.list_files("/home/daytona"):
    if file.endswith('.png'):
        sandbox.download_file(file, f"./{file}")
        print(f"✓ Chart saved to {file}")

# Summarize results
summary = client.chat.completions.create(
    model=SUMMARY_MODEL,
    messages=[{
        "role": "user",
        "content": f"Summarize these analysis results: {result.stdout}"
    }]
)

print(summary.choices[0].message.content)

sandbox.delete()

import OpenAI from 'openai';
import { Daytona } from '@daytona/sdk';

const openai = new OpenAI();
const daytona = new Daytona({ apiKey: process.env.DAYTONA_API_KEY });

const CODING_MODEL = 'gpt-5.1';
const SUMMARY_MODEL = 'gpt-4o';

// Generate code
const codeResponse = await openai.chat.completions.create({
  model: CODING_MODEL,
  messages: [{
    role: 'user',
    content: `Analyze cafe sales data and create visualizations...`
  }]
});

// Execute in sandbox
const sandbox = await daytona.createSandbox();
await sandbox.uploadFile('data.csv', './cafe_sales_data.csv');
await sandbox.exec('pip install pandas matplotlib');
const result = await sandbox.exec(`python -c "${generatedCode}"`);

console.log('✓ Analysis complete');

import dspy
from daytona_interpreter import DaytonaInterpreter

# Configure DSPy
lm = dspy.LM("openrouter/google/gemini-3-flash-preview")
dspy.configure(lm=lm)

# Create RLM with Daytona interpreter
interpreter = DaytonaInterpreter(
    api_key=os.getenv("DAYTONA_API_KEY")
)

rlm = dspy.RLM(
    signature="question -> answer: str",
    interpreter=interpreter,
    verbose=True,
)

# RLM can call llm_query() from generated code
result = rlm(
    question="""Analyze sales trends and use llm_query() to 
                generate insights for each trend pattern."""
)

print(result.answer)
interpreter.shutdown()

from daytona_interpreter import DaytonaInterpreter

def search_knowledge_base(query: str) -> str:
    """Custom tool that runs on host"""
    return json.dumps({"results": fetch_from_db(query)})

def query_api(endpoint: str) -> str:
    """Another custom tool"""
    return requests.get(endpoint).json()

# Pass tools to interpreter
interpreter = DaytonaInterpreter(
    tools={
        "search_knowledge_base": search_knowledge_base,
        "query_api": query_api,
    }
)

# LLM can now call these functions in generated code
rlm = dspy.RLM(
    signature="question -> answer",
    interpreter=interpreter
)

# Provide clear data descriptions
tool.upload_file(
    file=dataset,
    description="""Clear description of:
    - Column names and types
    - Data quality issues
    - Required cleaning steps
    - Business context
    """
)

try:
    sandbox = create_sandbox()
    result = run_analysis(sandbox)
finally:
    sandbox.delete()  # Always cleanup

def process_results(artifacts):
    """Custom handler for analysis outputs"""
    charts = [a for a in artifacts if a.endswith('.png')]
    tables = [a for a in artifacts if a.endswith('.csv')]
    
    return {
        'visualizations': charts,
        'data_exports': tables
    }

# Install required packages before analysis
tool.install_python_packages([
    'pandas',
    'matplotlib',
    'seaborn',
    'scikit-learn'
])

result = agent.invoke({
    "input": """Perform EDA on the dataset:
    1. Show summary statistics
    2. Identify missing values
    3. Detect outliers
    4. Show correlation heatmap
    """
})

result = agent.invoke({
    "input": """Analyze sales over time:
    1. Identify trends and seasonality
    2. Forecast next quarter
    3. Highlight anomalies
    """
})

result = agent.invoke({
    "input": """Build predictive model:
    1. Feature engineering
    2. Train/test split
    3. Model selection
    4. Evaluate performance
    """
})

from langchain_daytona_data_analysis import DaytonaDataAnalysisTool

tool = DaytonaDataAnalysisTool(api_key: str)

# Upload dataset
tool.upload_file(file: IO, description: str) -> SandboxUploadedFile

# Download results
tool.download_file(remote_path: str) -> bytes

# Install packages
tool.install_python_packages(packages: str | list[str]) -> None

# Cleanup
tool.close() -> None