T his playbook is your guide to building a fully AI-powered analytics system that transforms raw business data into fast, actionable insights. By leveraging a modular “AI Data Science Team” of SQL and visualization agents, this repo makes it easy to:
- Query your business database in plain English, no SQL needed
- Get instant answers, charts, and summaries from GPT-4o and LangChain agents
- Visualize and explain your data with flexible, plug-and-play AI modules
- Run the whole workflow in your own secure cloud, using Flask, Streamlit, and AWS
🌐 Live Architecture
- EC2: Flask backend with Gunicorn and Nginx
- S3: Secure file storage for datasets
- RDS (MySQL): Stores metadata and session logs
- Lambda: Triggers OpenAI agents for data cleaning
- Secrets Manager: Secure API key management
- GitHub Actions: Handles CI/CD deployment
Backend
Flask
Python
Gunicorn
Frontend
HTML
Streamlit
Jinja2
TailwindCSS
AI Agents
GPT-4o
AI Data Science Team
LangChain
Infrastructure
S3
EC2
RDS
Lambda
Secrets Manager
CI / CD
GitHub Actions
Each chapter guides you through designing and wiring up modular agents—so you can ask complex business questions in plain English and get back SQL, clean data, or beautiful visualizations instantly. Every step of the pipeline for your own business needs.
✅ AI Agent Task Force
Data Visualization
Business SQL Agent
Data Wrangling
Feature Engineering
📚 Playbook Chapters
1. ML Lead Scoring (H2O)
Kick off with automated machine learning: train and deploy an H2O AutoML model to score your leads, helping sales and marketing prioritize with real predictive power.
2. Natural Language SQL Agent
Use LangChain + OpenAI to ask questions in plain English and get optimized SQL queries with real answers—no manual SQL needed.
3. Orchestrating Agents with LangGraph
Build modular workflows as Directed Acyclic Graphs (DAGs), connecting agents together so your system is transparent, debuggable, and scalable.
4. From SQL to DataFrames and Visualization
Teach your agents to transform SQL results into Pandas DataFrames, and automatically decide when to display a chart or a table.
5. Charting Agent (Plotly + AI)
Let AI write and execute Python code that turns data into beautiful, interactive Plotly charts—on the fly, based on the user’s intent.
6. Modularization & Production
Wrap up all agents and workflows into reusable Python modules—plug and play for any future data pipeline or product.
7. Live Copilot Dashboard
Deliver a user-friendly, live dashboard with Streamlit—powered by your agentic backend, with authentication, history, and live analytics.
8. Multi-Agent Teamwork
See how specialized agents (SQL, EDA, Charting) work as a team—ready to be extended with more business logic, analytics, or LLM-powered features.
"This is super cool! I remember struggling to bridge that gap between insights and code. A tool that shows the 'how' is a total game-changer for learning." - Tran Tien Van
📁 Repository Structure
sql_agent_orchestration/
│
├── 01_machine_learning_h2o.py
├── 02_sql_agent.py
├── 03_sql_agent_langgraph.py
├── 04_add_pandas_langgraph.py
├── 05_add_routing_langgraph.py
├── 06_add_plotting_langgraph.py
├── 07_modular_bi_agent.py
├── 08_streamlit_app_persistant_plot_and_df.py
├── 09_streamlit_app_bi_copilot.py
├── 10_ai_data_science_team.py
│
├── additional-requirements.txt
├── credentials.yml
├── README.md
│
├── business_intelligence_agent/ # BI agent Python package
├── database/
│ ├── leads_scored.db # SQLite database
│ ├── leads_scored.db-shm
│ └── leads_scored.db-wal
├── models/ # ML models
├── notebooks/ # Jupyter notebooks
└── project2_architecture.drawio # Project architecture diagram
📁 Run Locally
git clone https://github.com/LannonTheCannon/Cloudberry_AWS_Bootcamp.git
cd Cloudberry_AWS_Bootcamp/Portfolio_V2
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python3 app.py