From 787e302ee6638d4ccf6c5a369348e1b0ba7af743 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Thiberg?=
 <42668688+bjornthiberg@users.noreply.github.com>
Date: Mon, 9 Sep 2024 21:46:14 +0200
Subject: [PATCH] Added proposal demo (#2452)

proposal demo for week 4, mlflow
---
 .../demo/week4/bthiberg-omidfm/README.md      | 36 +++++++++++++++++++
 1 file changed, 36 insertions(+)
 create mode 100644 contributions/demo/week4/bthiberg-omidfm/README.md

diff --git a/contributions/demo/week4/bthiberg-omidfm/README.md b/contributions/demo/week4/bthiberg-omidfm/README.md
new file mode 100644
index 0000000000..c0c4222dba
--- /dev/null
+++ b/contributions/demo/week4/bthiberg-omidfm/README.md
@@ -0,0 +1,36 @@
+# Assignment Proposal
+
+## Title
+
+Dynamic model rollbacks using MLflow
+
+## Names and KTH ID
+
+  - Björn Thiberg (bthiberg@kth.se)
+  - Omid Fattahi Mehr (omidfm@kth.se)
+
+## Deadline
+
+- Week 4
+
+## Category
+
+- Demo
+
+## Description
+
+We will demonstrate how to implement and automate dynamic model rollbacks using MLflow’s Model Registry. 
+
+We will look at a model in production, and simulate a model performance drop or condition (set by us) when deploying a new version of the model.
+
+Based on this condition, we will then trigger a rollback to an earlier reliable model (using Python code and the MLFlow API).
+
+The process will be verified and visualized using the MLflow GUI.
+
+**Relevance**
+
+Using version management, automatic deployment and automatic reactions based on feedback from the production environment are core parts of the devops philosophy for software development. 
+
+By doing this in an MLOps context, we demonstrate how DevOps principles can be extended for use in MLOps.
+
+There is no way to natively do this yet in MLflow, and there is even currently [an issue](https://github.com/mlflow/mlflow/issues/6180) on the main MLflow repository, showing that this is both difficult and relevant.