FlowForge AI

Agentic MLflow Model Development & Governance Platform

FlowForge AI is an open-source, containerized Machine Learning platform that automates the end-to-end ML lifecycle using AI Agents. It features a Human-in-the-Loop workflow, ensuring you stay in control while AI handles the heavy lifting of EDA, Feature Engineering, and Model Optimization.

🏗️ Architecture

The system is composed of several Docker services orchestrated via Docker Compose:

graph TD
    User((User)) -->|Browser| UI[Streamlit UI : 8501]
    UI -->|HTTP| API[FastAPI Backend : 8000]
    API -->|Manage| Prefect[Prefect Orchestrator : 4200]
    API -->|Log| MLflow[MLflow Tracking : 5000]
    API -->|Prompt| Ollama[Ollama LLM GPU : 11434]
    Prefect -->|Run| Workers[ML Workers]
    Workers -->|Train| Optuna[Optuna Optimization]
    Workers -->|Track| MLflow
    API -->|RAG| Artifacts[File System /app/docs]

Loading

Key Components

• Streamlit UI: Interactive Wizard-style interface.
• FastAPI Backend: Handles requests, runs Agents, and manages file I/O.
• Prefect: Orchestrates long-running flows (Model Training).
• MLflow: Tracks experiments, metrics, and models.
• Ollama: Local LLM inference (Llama 3 recommended) for Agents.
• Dozzle: Real-time log viewer.

🚀 Getting Started

Prerequisites

• Docker and Docker Compose installed.
• NVIDIA GPU & NVIDIA Container Toolkit (for Ollama acceleration).
  • Note: If you don't have a GPU, modify docker-compose.yml to remove the deploy section under ollama.

Installation

1. Clone the Repository

   git clone https://github.com/quickSilverShanks/FlowForge-AI.git
   cd FlowForge-AI

2. Start the Platform

   docker-compose up --build -d

   The first run will take time to build images.

3. Initialize Ollama Model Once containers are running, download the LLM model (execute inside the ollama container or via API if implemented, typical manual step for first time):

   docker exec -it flowforge_ollama ollama pull llama3

Configuration

You can configure the LLM model used by the agents by setting the LLM_MODEL_NAME environment variable in the .env file at the root of the project.

1. Create a .env file (if not present):

   cp .env.example .env

2. Edit .env and set your desired model:

   LLM_MODEL_NAME=llama3

   Note: Ensure the model you choose strictly matches the model name in Ollama.

Usage Guide

Access the services at the following URLs:

Service	URL	Description
App UI	http://localhost:8501	Main Interface (Streamlit)
Backend API	http://localhost:8000/docs	API Swagger
MLflow	http://localhost:5000	Experiment Tracking
Prefect	http://localhost:4200	Workflow Orchestration
Logs	http://localhost:8888	Container Logs

🧪 Workflow Steps

1. Data Upload: Go to the "Data Upload" page and upload your CSV/Parquet file. Define your Target column.
2. EDA: Switch to "EDA". Click "Run AI Analysis". The Agent will generate a "Vibe Check" and statistical summary.
3. Feature Engineering: Go to "Feature Engineering". The Agent will propose a plan (Imputation, Encoding, etc.). Review, Edit, and Apply it.
4. Model Training: Go to "Model Training". The Agent proposes a search space. Click "Start Training" to trigger the Prefect Flow. Watch the progress in the Prefect Dashboard.
5. Validation: Evaluate fairness and performance on OOT data.
6. Monitoring: Generate a drift detection config.
7. Final Report: Chat with the RAG Agent to ask questions about what happened during the session (e.g., "Why did we drop the Age column?").

🛠️ Development

• Backend Code: app/api/
• UI Code: app/ui/
• Agents: app/core/agents/
• ML Logic: app/core/ml/

To add a new dependency, update requirements.txt and rebuild:

docker-compose up --build -d