InsightData is an advanced AI Data Analyst powered by Google's Gemini 2.5 Flash and built upon the ReAct (Reasoning + Acting) architecture.

Unlike standard chatbots, InsightData doesn't just guess answers. It actively reasons about your data—whether it's a local CSV/Excel file, a live Google Sheets link, or a Direct CSV URL. It functions like a real data scientist: understanding your dataset structure, writing and executing Python code (Pandas/Matplotlib) in real-time, cleaning data on-the-fly, and visualizing trends—all in your preferred language (English or Indonesian).

It enforces strict Deep Dive Logic: when asked "Why?", it refuses to rely on assumptions. Instead, it investigates statistical correlations (e.g., Price vs. Rating) to prove the reasons behind a trend.

Using the LangChain Pandas Agent framework, the assistant follows a strict cognitive loop:

1. Thought: Analyzes the user's question and plans the analysis steps.
2. Action: Generates and runs actual Python/Pandas code.
3. Observation: Reviews the execution result (numbers or plots) to ensure accuracy.
4. Final Answer: Synthesizes the findings into business insights.

• Dual Language Support: Seamlessly switch between English and Indonesian via the sidebar.
• Smart Translation: The agent enforces a strict logic layer to ensure the final answer matches your chosen language, regardless of the language used in the query.

• Generative Plotting: Just ask "Show me the sales trend" or "Plot the top 5 [Category Name]" (e.g., "Plot the top 5 most sold products"), and the AI will generate Matplotlib/Seaborn charts instantly.
• Interactive Downloads: Every generated plot comes with a direct 📥 Download Button so you can save insights for your reports immediately.

• Auto-Cleaning: Automatically removes "ghost" data (empty rows/cols) and duplicates upon upload to prevent system crashes.
• Smart Mode Switching: Prevents cross-contamination by automatically flushing memory when switching between Local Files and Cloud URLs.
• Hybrid Data Source: Supports local Files (.csv/.xlsx) and Cloud URLs (Google Sheets & Direct .csv links).
• Privacy First: Data is processed in temporary memory. The Agent cannot overwrite or delete your original files.

• LLM: Google Gemini 2.5 Flash (via ChatGoogleGenerativeAI).
• Framework: Streamlit (Frontend).
• Orchestration: LangChain (Pandas DataFrame Agent).
• Data Processing: Pandas, Matplotlib, Seaborn.
• Memory: ConversationBufferMemory.

To ensure the AI understands your data, your file MUST follow these rules:

• 🚫 No Merged Cells: Unmerge all cells.
• 📝 Single-Row Header: The first row must be the header.
• 🔢 Consistent Data Types: Do not mix text and numbers in one column (e.g., avoid 1,000 as string vs 1000 as int).

• Priority: The system prioritizes Local Files over URLs.
• Switching: If you want to analyze a URL, you must remove all uploaded files first.

1. Clone the Repository

   git clone https://github.com/viochris/insight-data-ai-analyst.git
   cd insight-data-ai-analyst

2. Install Dependencies

   pip install -r requirements.txt

3. Run the Application

   streamlit run app.py

1. Configuration (Sidebar):
   • Enter your Google Gemini API Key.
   • Select your Language Preference (English/Indonesian).
2. Load Data:
   • Option A: Upload .csv or .xlsx files.
   • Option B: Paste a public Google Sheets URL OR a direct .csv link.
   • Click "⚡ Initialize & Analyze" to wake up the Agent.
3. Query:
   • Type your question naturally (e.g., "What is the total revenue?" or "Visualize the correlation between price and rating").
   • Watch the "Thinking Process" expander to see the Python code being generated.
4. Manage:
   • Use "🧹 Clear Conversation" to tidy up the chat while keeping data loaded.
   • Use "🔴 Hard Reset System" to wipe everything and restart the session.

The clean, modern landing page offering quick configuration for API keys, data uploads, and language settings.

The Agent generates standalone, high-resolution matplotlib charts that can be downloaded directly for reports.

The ReAct engine provides a detailed breakdown of the analysis. Below is the complete flow captured in three parts:

Part 1: Query & Response Generation

The Agent receives the natural language prompt, instantly generates the visualization within the chat, and begins the analysis.

Part 2: Deep Dive Analysis

Instead of just describing the chart, the Agent calculates supporting metrics (rating, price, recommendation rate) to explain why the trend exists (e.g., finding that 'Casual' style is popular due to lower average price).

Part 3: Final Insight

The AI synthesizes all statistical findings into a clear, business-friendly summary in the user's chosen language.

Author: Silvio Christian, Joe "Stop writing complex Python code. Start asking questions."