🧪 Automated Statistical Hypothesis Testing Agent

An agentic, "best-in-class" statistical analysis system built with Agno, Azure OpenAI, and a high-performance data backbone (Polars/Dask/Pandas). This project demonstrates how AI agents can automate rigorous scientific research workflows, from data profiling to executive-level reporting.

Note

This project is a Proof of Concept (POC). It represents a foundational idea for an autonomous researcher and can be expanded for specific domain-expert applications (e.g., Clinical Trials, Market Research, A/B Testing at Scale).

---

🌟 Why This Project?

Traditional statistical software (SPSS, SAS, R) either requires deep technical expertise or is too rigid for rapid business inquiry. This project was built to bridge that gap:

1. Lower the Barrier: Allow anyone to ask complex business questions in natural language.
2. Ensure Rigor: Automate the "boring but critical" parts of statistics (normality checks, variance testing, sample size calculation) to prevent common human errors.
3. Handle Big Data: Enable analyzing millions of rows on a laptop without specialized infrastructure.

🛠️ How It Works (The Core Engine)

1. Multi-Agent Orchestration (Agno Team)

Powered by a specialized Agno Team, the system divides the workload:

• 🔍 Data Analyst: Proactively profiles the schema and identifies the "Ground Truth" columns needed.
• ⚖️ Stats Executor: The "Brain" that chooses and runs the right test (T-Test, ANOVA, Chi-Square, Mann-Whitney) only after validating mathematical assumptions (e.g., Shapiro-Wilk for normality).
• 📝 Insights Reporter: Translates the "Greek and Math" into a professional business verdict with actionable recommendations.

2. Dynamic Statistical Sampling (Cochran’s Theorem)

Instead of arbitrary limits, the system uses a Dynamic Sampling Engine:

• Formula-Driven: Uses Cochran’s Formula (99.9% Confidence, 0.5% Margin of Error) to calculate the exact sample size needed for a representative population slice.
• Big Data Ready: Seamlessly handles files up to 2GB using Dask and Polars while maintaining UI responsiveness.

3. Robustness Engineering

Every statistical tool is "hardened" with a validation layer:

• Defensive Programming: Detects empty groups, zero-variance data, and insufficient samples before execution.
• Self-Healing: Provides human-readable error messages to the agent, allowing it to pivot or explain data limitations.

🏗️ Technical Stack

• Agentic Framework: Agno
• LLM Backend: Azure OpenAI (GPT-4o)
• Data Processing: Polars (Fast), Dask (Parallel/Large), Pandas (Standard)
• Scientific Suite: Scipy, Statsmodels, Pingouin (Modern Stats)
• UI Foundation: Streamlit (Premium Custom CSS)
• Environment: Python 3.12 + uv package manager

🚀 Getting Started

Prerequisites

• Python 3.12
• Azure OpenAI Deployment (GPT-4o recommended)

Setup

1. Clone the repository.
2. Create a .env file with your credentials:

   AZURE_OPENAI_API_KEY=your_key
   AZURE_OPENAI_ENDPOINT=your_endpoint
   AZURE_OPENAI_API_VERSION=2024-05-01-preview
   AZURE_OPENAI_DEPLOYMENT=your_deployment_name

3. Install dependencies:

   uv pip install -r requirements.txt

4. Run the app:

   streamlit run src/app.py

👨‍💻 Who Is It For?

• Business Leaders: Get instant, scientifically valid answers to "what-if" questions.
• Data Teams: Automate the initial "sanity check" phase of research.
• AI Researchers: Explore agentic patterns for complex scientific tool-calling.

📈 Future Potential

This POC can be expanded into a "Researcher-in-a-Box":

• Automated Visualization: Dynamically generating charts based on the result of the test.
• Correlation Matrix Exploration: Autonomously identifying high-impact variables.
• Advanced ML Integration: Predicting future trends based on the discovered hypothesis.

---

Built with ❤️ for Robust Statistical Autonomy.