continuous-training

🧹 Automated Filtering of Undesirable Web Data to Update LLM Knowledge

Open-Source Machine Learning Platform

Data version control with Makefile and DVC for a regression task to estimate insurance costs for certain individuals.

🏷️ An AI-driven approach to Label LLM Training Data

mini project

The runtime environment for the KOI-System. Train models, run instances, and collect samples.

The backend of the KOI-system.

Fully automated MLOps pipeline with continuous training, drift detection, model versioning, and self-healing deployments using canary releases and real-time monitoring.

🌱 Manifeste de la Clairveillance : pour des institutions qui mesurent, doutent et apprennent

A production-grade, end-to-end MLOps platform designed to automate the complete lifecycle of a financial fraud detection system. This repository demonstrates how to transition from a static ML model to a self-healing, automated production system.

End-to-end Financial MLOps pipeline for automated stock forecasting featuring PyTorch LSTM, Apache Airflow orchestration, MLflow tracking, MongoDB Atlas, and FastAPI containerized with Docker and GHCR.

Closed-loop MLOps control plane: versioned data, gated training, canary releases, drift detection, automated retraining and rollback.

Great Expectations gating data at the pull request, DVC versioning datasets in S3, and models landing as PendingManualApproval behind an IAM action CI does not hold: continuous training with a human in the loop. A Go CLI drives SageMaker training and the Model Registry from GitHub Actions over OIDC, on Terraform with no idle compute.

A concurrent training and generation pipeline leveraging active learning to drive synthetic data rendering. By generating customized datasets simultaneously alongside model training, it creates a real-time feedback loop to dynamically refine object detection models.

This project integrates Airflow, EC2, MLFlow, and MLOps principles to deploy a robust pipeline for classifying medical MNIST images. Streamlit enables user-friendly image uploads, MLFlow handles model registry and inference, while Airflow automates data updates and model retraining on AWS EC2.

Closed-loop MLOps pipeline for demand forecasting: FastAPI serving with prediction logging, Airflow monitoring a 14-day rolling RMSE to trigger retraining, leak-free chronological validation, and a promotion audit so a new model ships only if it beats production.

Enterprise-grade Self-Healing MLOps Pipeline with Hybrid Decision Engine (Rules + Contextual Bandits), Automated Drift Monitoring, Airflow, and K8s Zero-Downtime Rollouts