README.md

metadata

title: MLOps AI Agent
emoji: ♻️
colorFrom: indigo
colorTo: green
sdk: gradio
sdk_version: 6.0.0
app_file: app.py
pinned: false
license: apache-2.0
short_description: Analyse csv, train, deploy, and get endpoint API.
tags:
  - mcp-in-action-track-enterprise

🧠 MLOps Agent: Auto Analysis, Training & Deployment

A fully automated AI powered MLOps agent that you can run in a browser: Upload a CSV → Auto-analyze → Train a model → Deploy → Get example inference code.

This app uses:

• Gradio (streaming UI)
• OpenAI Responses API (gpt-5-mini / gpt-5)
• Remote MCP server for tool calling
• CSV→Model→Deployment pipeline
• Streaming chat UX with collapsible technical details
• Automatic Python & curl example code generation

Link to the LinkedIn post: https://www.linkedin.com/feed/update/urn:li:ugcPost:7399748605563277312/

Link to the YouTube video: https://www.youtube.com/watch?v=dzVgoG1uk0M

🚀 Features

📋 File Size Limits

• Hard cap: 10MB maximum file size
• Soft cap: 1.5MB recommended for optimal performance
• Large file handling: If a file exceeds the soft cap, training will automatically use only the top 10K rows for processing
• This optimization speeds up the training process and returns model IDs and responses faster

1. Natural Chat Interface

Chat like a normal assistant — ask questions, discuss data, clarify tasks.

2. Automatic CSV Analysis

Upload any structured CSV and ask:

“Analyze this dataset”

The agent will produce:

• Data shape
• Target detection
• Basic stats
• Data quality issues
• Recommendations

3. Model Training

Ask:

“Train a model on this dataset”

The agent will automatically:

• Train a classifier or regressor
• Evaluate metrics (Accuracy, F1, ROC-AUC…)
• Summarize key performance signals

4. Automated Deployment

Ask:

“Deploy the model”

The MCP backend handles deployment and returns:

• Deployment status
• Example Python inference code
• Example curl usage

5. Clean, Structured Responses

Every response includes:

Key Summary → Example Usage → Technical Details (collapsible)

• Main explanation in clear Markdown
• Python + curl examples outside the collapsible
• Tools, parameters, and logs inside a <details> block

6. Full Streaming Output

All assistant messages stream token-by-token for fast UX.

🖥️ UI Overview

Chat + File Upload

• Drag and drop a CSV
• Type queries (analysis, training, deployment, follow-ups)
• Live streaming assistant responses

Collapsible Technical Details

The agent always displays a <details> block containing:

• Tool names used
• Parameters passed
• Training logs
• Deployment logs
• Raw JSON (if available)

This keeps the main summary clean while still exposing full transparency.

Clickable Examples

• Pre-loaded examples for quick testing
• Triggers analysis, training, deployment, and auto-deployment
• Uses the included data/heart.csv and data/housing.csv dataset automatically

📦 Installation

Clone the repo:

git clone https://huggingface.co/spaces/MCP-1st-Birthday/mlops-agent
cd mlops-agent

Install dependencies:

pip install -r requirements.txt

Set your environment variable:

export OPENAI_API_KEY="your-key-here"

Run the app:

python app.py

🧬 Architecture

User → Gradio Chat UI
      → OpenAI Responses API (streaming)
          → MCP Server (tools: analyze, train, deploy)
              → Outputs come back as structured signals
→ Assistant formats them using strict markdown schema
→ Gradio displays stream + collapsible logs

🛠️ Tech Stack

• Python 3.10+
• Gradio 6.x
• OpenAI SDK
• Remote MCP Tool Server (auto-train, evaluate, deploy)
• Streaming I/O architecture