VizzBreeze

A lightweight, high-performance transactional web-dashboard designed for advanced data flow analytics, multi-stage routing visualization, and statistical risk audit，and AI Agent trace visualization. Built natively on top of Streamlit, Pandas, and Plotly.

Key Features

• Flat Transaction Processing Engine: Optimized for fast processing of un-aggregated logs (e.g., Client ➔ Fund ➔ Asset).
• Global Session Synchronization: Seamlessly locks metric column selections, typography sizes, and alignments across all workspace views.
• Auto-Scroll Suppression: Enhanced layout architecture prevents viewport jumps during widget updates and data updates.
• Advanced Graph Matrix Analytics: Includes high-fidelity stacked charts, multi-dimensional Parcats layouts, Density Matrices, and automated IQR risk audit profiling tools.
• AI Agent Trace & Observability: Native templates designed to map LLM chain-of-thought routing, monitor token consumption footprints, and catch execution latency anomalies.

Build With

1 - Plotly - Core interactive charting engine. 2 - Streamlit - Cloud infrastructure and web UI framework. 3 - Pandas - High-performance data structures and data analysis engine.

Quick Start

1. Standard Python / Jupyter Notebook Usage

VizzBreeze functions process un-aggregated raw DataFrames and return native Plotly figures, making them fully compatible with Jupyter views and pipeline automation:

import vizzbreeze as vb
import pandas as pd

# Load raw transaction logs
df = pd.read_csv("transactions.csv")

# Generate advanced Parcats flow directly in your notebook
fig = vb.generate_parcats(
    df=df, 
    stage_nodes=['client', 'fund'], 
    target_node='asset', 
    value_col='amount'
)
fig.show()

2. Standalone Web Control Room Execution

If you prefer a full-scale interactive UI with global layout sync, just open your local terminal and run:

vizzbreeze-run

Cloud Execution

If you explicitly need to run VizzBreeze in cloud environments like Google Colab, please use a secure SSH/Port-forwarding tunnel to bypass iframe infrastructure limits.

Test Dataset

To explore the dashboard features instantly, you can use the pre-configured spreadsheet sample_data_unaggregated.xlsx located in the root folder of this repository. Just drag and drop it into the sidebar upload zone!

Accessing Built-in Color Palettes

You don't need to hardcode HEX styles. Access the corporate design system directly from the package configuration:

import vizzbreeze as vb

# View all available palette names
print(vb.COLOR_PALETTES.keys())

# Extract a specific synchronized list of colors
chosen_colors = vb.COLOR_PALETTES["Warm Amber"]

Core Analytical Tabs

1. Flows: Explore multi-stage category paths (e.g., LLM chain-of-thought: Intent ➔ Agent ➔ Tool) with thin, light axis labels and custom high-contrast hover tooltips.
2. Funnel: Track progressive conversion drops along specific workflow layers.
3. Structural Breakdown: Classic stacked column bar charts configured with responsive axis titles that update on the fly based on active filters.
4. Bento: An asymmetrical modular tile framework designed to compress complex multi-level dimensions (like Token Consumption Footprints) into a clean, prioritized grid dashboard.
5. Heatmap: High-density dashboard to scan cluster intersections instantly.
6. Anomaly & Risk Audit: Automated statistical anomaly profiling that detects execution latency spikes and infinite loop outliers without freezing the browser engine.

Core Functions API Reference & Parameter Mapping

All visualization engines are fully modular, accept un-aggregated raw pandas.DataFrame inputs, and return native plotly.graph_objects.Figure interactive objects.

1. Flows

flow_fig = vb.generate_parcats(
    df=df,
    stage_nodes=['user_intent', 'active_agent', 'tool_called'],
    target_node='execution_status',
    value_col='tokens_used',  # Flows will scale by token consumption!
    chart_title="AI Agent Chain of Thought & Token Distribution",
    selected_palette=chosen_colors,
    title_size=20,
    width_px=1600,
    height_px=500,
    title_x = 0.5
)

flow_fig.show()

2. Funnel

fig = vb.generate_funnel_chart(
    df=df,
    stage_nodes=['user_intent', 'active_agent', 'tool_called'],   
    target_node='execution_status',                               
    value_col='tokens_used',                                      
    
    # ИСПРАВЛЕНИЕ: Передаем параметры фильтрации из таблицы логов
    selected_route_dict={
        'user_intent':'Tech Support', 'active_agent':'Router_Agent', 'tool_called':'Knowledge_Base_Lookup'  # Показываем путь только для конкретного ID лога

    },                                       
    selected_palette=chosen_colors,            
    unit_divider=1.0,
    force_shuffle=True,
    chart_title="AI Agent Chain of Thought & Token Distribution", 
    width_px=1050,
    height_px=500,
    title_x=0.5
)

fig.update_layout(
    margin=dict(l=350, r=20, t=100, b=50)
)

3. Structural Breakdown

stage_nodes=['user_intent']
target_node='execution_status'
value_col='tokens_used'

fig = vb.generate_stacked_bar_chart(
    df=df,
    stage_nodes=stage_nodes,
    target_node=target_node,           
    value_col=value_col,            
    selected_palette=chosen_colors,
    unit_divider=1.0,
    force_shuffle=True,
    chart_title="Systemic Error Distribution Across Active Agents",
    width_px=1600,
    height_px=500,
    title_x = 0.5
)
fig.update_layout(
    margin=dict(l=100, r=20, t=50, b=120)
)

fig.update_layout(
    yaxis=dict(
        title=""
    )
)

fig.show()

4. Bento

fig = vb.generate_bento_treemap(
    df=df,
    id_col='user_intent',      
    value_col=value_col,           
    selected_palette=chosen_colors,
    unit_divider=1.0,
    force_shuffle=True,
    chart_title="Token Consumption Footprint by Intent Layer",
    width_px=1050,
    height_px=500,
    title_x = 0.5
)
fig.update_layout(
    margin=dict(l=50, r=20, t=50, b=50)
    )
fig.show()

5. Heatmap

fig = vb.generate_heatmap(
    df=df,
    x_col='user_intent',
    y_col='execution_status',
    value_col='tokens_used',
    selected_palette=chosen_colors,
    chart_title="AI Agent Token Consumption Heatmap by Execution Status", 
    show_annot=True,
    width_px=1050,
    height_px=500,
    title_x=0.5
)

fig.update_xaxes(title_text="USER INTENT LAYER")
fig.update_yaxes(title_text="EXECUTION STATUS")

fig.update_layout(
    margin=dict(l=150, r=20, t=50, b=50)
)
fig.show()

6. Anomaly & Risk Audit

anomaly_fig = vb.generate_outliers_chart(
    df=df,
    stage_col='active_agent',
    target_col='tool_called',
    value_col='execution_time_sec',
    agg_func='max', 
    chart_title="AI Execution Latency Anomaly Profile (IQR Bounds)",
    selected_palette=chosen_colors,
    width_px=1050,
    height_px=500,
    title_x=0.5
)

anomaly_fig.update_xaxes(title_text="MAX SINGLE-RUN EXECUTION TIME (SEC)")

anomaly_fig.update_layout(
    margin=dict(l=310, r=20, t=50, b=50)
)
anomaly_fig.show()

Requirements

• Python >= 3.9
• Streamlit >= 1.35.0
• Pandas >= 2.1.0
• Plotly >= 5.18.0

Roadmap

• Advanced Data Preprocessing Pipeline: Implementing automated anomaly tracking and missing value imputation optimized for multi-source financial and operational data arrays.
• Interactive Risk Metrics: Adding features for deep data analysis, including volatility tracking, anomaly detection, and custom structural breakdown widgets.
• Large-Scale Data Engineering: Adding chunks-based data loading and optimization widgets to seamlessly process unaggregated corporate and industrial datasets exceeding 500MB.
• Enterprise Stability & Testing: Expanding code verification with an extensive pytest suite to achieve 85%+ test coverage for secure local CLI execution.

License

This project is licensed under the MIT License - see the LICENSE file for details.