ModelLensAI - Model Analysis Platform

A production-oriented Flask application for uploading ML and DL model artifacts, benchmarking supported runtimes, logging MLOps metadata, serving inference, monitoring latency, and managing deployment stages.

What this build now supports

• classical ML artifacts such as joblib, pickle, and serialized sklearn-family models
• TensorFlow and Keras .keras and .h5 models
• TensorFlow Lite .tflite
• TensorFlow SavedModel packaged as .zip
• PyTorch TorchScript .pt
• PyTorch checkpoints with structural inspection
• Ultralytics YOLO .pt/.pth image inference when the runtime library is installed
• ONNX runtime benchmarking and live inference
• live prediction from the report page for supported feature and image models
• MLflow support with JSON fallback logging
• rate limiting, readiness checks, CLI bootstrapping, and deployment stages

Local run

python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements-lite.txt
# For full TensorFlow support, use Python 3.9-3.12 and install requirements.txt instead.
flask --app run.py init-db
flask --app run.py create-demo-user
python run.py

Open http://127.0.0.1:5000

Demo account:

• username: demo
• password: demo123

Production notes

• use PostgreSQL instead of SQLite
• use Redis-backed rate limiting storage
• put the app behind Nginx or another reverse proxy
• store uploads in object storage if artifacts are large
• keep secrets in environment variables or a secret manager
• prefer portable runtime exports for deployment analysis: ONNX, TorchScript, Keras, or TFLite

Environment variables

• FLASK_ENV=development|production
• AUTO_INIT_DB=true|false
• DATABASE_URL=sqlite:///... or PostgreSQL URL
• RATELIMIT_STORAGE_URI=redis://localhost:6379/0
• SESSION_COOKIE_SECURE=true|false
• REMEMBER_COOKIE_SECURE=true|false
• TRUST_PROXY=true|false
• MLFLOW_EXPERIMENT_NAME=model-analysis-platform
• MAX_CONTENT_LENGTH_MB=512

Useful commands

flask --app run.py init-db
flask --app run.py reset-db
flask --app run.py create-demo-user

Supported runtime stacks

• Machine learning: scikit-learn, XGBoost, LightGBM, CatBoost, skops, joblib/pickle loaders
• Deep learning: PyTorch, TorchVision, TensorFlow, Keras, TFLite, ONNX Runtime, Ultralytics YOLO
• Data science: NumPy, Pandas, PyArrow, SciPy, Matplotlib, Seaborn, Plotly, SHAP

UI

• Enterprise-style upload workflow
• Registry view with framework badges
• More detailed deployment report layout
• Module visibility for ML, DL, data science, and Ultralytics support