Architecture Overview

The system is built with modularity and delegation in mind. Each AI agent is responsible for a specific part of the workflow:

Triaging Agent: Interprets the user query and breaks it into tasks.

Cleaning Agent: Removes duplicates and handles missing values.

Statistical Agent: Calculates descriptive stats.

Visualization Agent: Generates line chart data.

Correlation Agent: Measures relationships between variables.

These agents are orchestrated using a central execution handler, which ensures that the output of one agent can serve as input to the next.

The Multi-Agent Data Analysis System is composed of purpose-built agents, each responsible for a distinct subset of tasks within the data analysis pipeline. These agents leverage OpenAI’s GPT-4 architecture to deliver context-aware, intelligent task execution.

Triaging Agent

triaging_agent = TriagingAgent(OPENAI_MODEL)

    conversation_history = handle_user_message(user_query, triaging_agent)

The Triaging Agent serves as the system's entry point, responsible for parsing and interpreting natural language user queries.

Key Responsibilities:

• Perform semantic analysis on user input to identify intent

• Route requests to appropriate downstream agents

• Maintain conversational context and manage multi-turn interactions

• Request clarification or additional parameters if input is ambiguous

• Provide graceful fallback and error-handling mechanisms

Example Interaction:

User: “I need to analyze sales data trends.”

Triaging Agent: “Got it. To get started, could you specify:

1. The time period you're interested in

2. The specific metrics you want to analyze

3. Any preferred type of visualization?”

Data Processing Agent

The Data Processing Agent handles all stages of data wrangling and preparation, built on top of pandas and numpy.

Key Responsibilities:

• Data Cleaning

• Remove duplicates

• Handle missing values (e.g., imputation, deletion)

• Standardize column formats

• Detect and process outliers

Data Transformation:

• Feature scaling (normalization, standardization)

• Encode categorical variables

• Parse and process date/time fields

• Apply domain-specific transformation logic

Aggregation:

• Perform grouped aggregations (e.g., sum, mean, median)

• Support multi-level grouping

• Apply time-windowed aggregation for temporal analysis

Analysis Agent

This agent performs quantitative analysis on structured data.

Key Responsibilities:

• Conduct descriptive statistical analysis (mean, median, variance, etc.)

• Calculate correlation coefficients (Pearson, Spearman)

• Execute linear and logistic regression analysis

• Interpret analysis results in the context of user goals

Visualization Agent

The Visualization Agent is responsible for converting data insights into interpretable visual formats.

Key Responsibilities:

• Generate bar, line, and pie charts

• Dynamically select the best-fit chart based on data type and user intent

• Return chart-ready data or render-ready JSON for front-end visualization pipelines

Analysis Tools Overview:

Processed Data:

Statistical Analysis:

Basic Stats: Mean/Median/Mode

Distribution: Standard Deviation

Hypothesis: T-tests/ANOVA

Correlation Analysis:

Pearson: Linear Correlation

Spearman: Rank Correlation

Regression Analysis:

Linear: Simple/Multiple

Logistic: Binary/Multi-class

Visualization Tools Components:

Analysis Results:

Chart Generation:

Bar Charts: Simple Bar, Grouped Bar, Stacked Bar

Line Charts: Simple Line, Multi-line, Area Chart

Pie Charts: Simple Pie, Donut Chart, Exploded Pie