Machine Learning Systems for Production (Batch-2)

This repository contains materials for the "Machine Learning Systems for Production" course by Thar Htet San. The lecture videos are available on the Deepfaro YouTube channel. This course focuses on MLOps, covering the following key topics:

• Addressing technical debts in MLOps.
• Framing problems as ML solutions.
• Approaching ML systems using current project management techniques.
• Setting up a development environment for ML projects.
• Creating an ML lifecycle management system.
• Handling data drift, model drift, and monitoring in ML systems.
• Systematically managing the ongoing development of ML systems.

Course Outlines

Chapter 1: Basic Python Logic

Day	Topics Covered	Source Code
Day 1	- Course Introduction - Tools and Git Config - Anaconda and Python Syntax	1_Python_Basic_and_git
Day 2	- Python class and object - Ollama Usage - Pandas - List - Function Decorator - Args vs kwargs	1_Python_Basic_and_git

Chapter-2 : Environment Setup

Day	Topics Covered	Source Code
Day 3	- Anaconda Usage - PipEnv setup - Poetry for python Envs	2_ProdAndDev_Env_Setup

---

Chapter-3 : Image Processing

Day	Topics Covered	Source Code
Day 4	- Basic Image Processing Techniques. - BGR and RGB. - Image lib : OpenCV and Pillow. - Line detection Sample Project	- Basic Img Functions - Background/ Foreground extraction

---

Chapter-4 : Machine Learning

Day	Topics Covered	Source Code
Day 5	- OCR Project - Decision Tree - K-Means - Navie Bayes	- Img Processing for Invoic - OCR Project
Day 6	- Linear Regression coding - Decission Tree, Navie Bayes, K-Means Coding - Single perceptron (from scratch )	- Decision Tree and Navie Bayes : Framework - Decision Tree : Native - K-Means Clustering : Framework - K-Means Clustering : Native - Navie Bayes : Framework - Navie Bayes : Native - Linear Regression - Logistic Regression

---

Chapter-5 : Deep Learning

Day	Topics Covered	Source Code
Day 7	- Deep Learning - Artificial Neural Network (ANN) - Convolutional Neural Network (CNN) - How calculate no of params for DL model - Tensorflow GPU install	- ANN : Framework - ANN : Native
Day 8	- Env Setup for DL Proj - TF Data Generators	- TF generator : binary - TF generator : categorical
Day 10	- Custom Data Generator - Developing Custom TF model - Custom Training	- TF Custom generator - TF Custom Model - TF Custom Training
Day 11	- Knowledge Distillation - Teacher and Student Networks - Tensorflow best practices	- Teacher and Student Networks

---

Chapter-6 : Containerization And Deployment

Day	Topics Covered	Source Code
Day 12	- Docker - Flask - FastAPI	- Flask App - FastAPI App
Day 13	- Fastapi Schemas - Async and Sync	- Fastapi Schemas : Request/Response Model - Async and Sync
Day 14	- Fastapi Schemas - Async and Sync - Text to Audio - Text to Text	- Text to Audio / Text to Text
Day 15	- Design Pattern - Structural vs Behavioral design patterns - Structuring the project. - Code formatter and clean code creation. - Introduction to CloudRun	- Design Pattern Examples - ML_in_Prod_batch_2_proj1
Day 16	- Set CI/CD in Project for CloudRun	- Cloud Build ymal
Day 17	- Understanding artifact registry - CloudRun versioning - Run-time Variable in CI/CD	-
Day 18	- Github Action - Code Formatting - Code Quality Check - Unit Testing - Automation the Unit Tests	ci_sample.yml

---

Chapter-7 : Data Version Control in ML Projs

Day	Topics Covered	Source Code
Day 19	- Introduction to Data Version Control (DVC) - DVC setup for dev environment. - Introduction to MLflow.	- DVC Setup Configs - DVC Repo - Getting Start with MLflow

---

Chapter-8 : Experiments Tracking in ML

Day	Topics Covered	Source Code
Day 20	- Getting start with MLflow - Logging API in MLflow - MLflow client - Model management in MLflow	- Logging API in MLflow - MLflow client -Model management in Mlflow
Day 21	- Compute Resources - Eng-to-End MLflow	-End-to-End Mlflow
Day 22	- MLflow backend store setup - CloudSQL setup for MLflow	-
Day 23	- MLflow Prod Setup Explain	Secure-MLflow-Server-for-production

---

Chapter-9 : ML Systems Monitoring

Day	Topics Covered	Source Code
Day 24	- Concept Drift - Data Drift - Feature Drift - Label Drift - Sample drift detection logic - Introduction to EvidentlyAI	- Data Drift Sample -Monitoring with Evidentlyai
Day 25	- Introduction to Monitoring Project - Project setup with Prometheus and Grafana	Monitoring-with-Prometheus-and-Grafana
Day 26	- Introduction to Evidently Project. - Setting up Evidently with streamlit App and FastAPI	-Evidently Project Sample -Monitoring Proj with EvidentlyAI
Day 27	- Explain detail about Evidently Project.	Monitoring Proj with EvidentlyAI

---

Chapter-10 : PipeLine Orchestration

Day	Topics Covered	Source Code
Day 28	- Data Orchestration - Mage Pipeline vs Airflow Pipeline - Airflow System Explain - Role and User management in Airflow - Bitshift Operators	Bitshift Operator
Day 29	- Chain in DAGs -Tasks flow - Decorators in Airflow - Ariflow Variable vs System Variable - Xcom - Connections -File Sensor	- Airflow Variable vs System Variable -File Sensors -Python Sensors
Day 30	- Scheduler in Ariflow - GCP Bucket Sensors - Custom Airflow Server Setup	- GCP Bucket Sensors -Custom Airflow Server Setup

---

Chapter-11 : Measuring ML model's performance and Matrices

• I will add this chapter in next batch 🤓

Chapter-12 : Distributed Workload in ML

Day	Topics Covered	Source Code
Day 31	- Apache Spark - Spark System Explain - Spark Native vs PandasAPI on Spark - Memory and lazy execution explain	-Apache Spark
Day 32	- MLflow on Spark Cluster - MLflow Genric Flavor - Distributed Training on Spark Cluster	- Custom-Apache-Spark-Cluster-run-databricks-locally
Day 33	- MLflow model Generic Flavor - 2 - Hyperparameter tuning with Hyperopt - CPU vs GPU vs Distributed Training - Distributed Training with MLflow Generic Flavor	-

---

Chapter-13 : Advanced ai techniques

Day	Topics Covered	Source Code
Day 34	- Retrieval-Augmented Generation (RAGs) Explain - RAGs Application and Sample Codes - RAGs vs AI Agent Systems	-RAG -Agentic RAG
Day 35	- Agentic RAGs with Ollama - Introduction to model context protocol (MCP) - Hello world in MCP	-Hello-world MCP
Day 36	- Tools calling in AI Agents - fuzzy output handling in agents - MCP with Claude	-Tools calling in AI Agents -MCP samples
Day 37	- Fuzzy Output Handling -Handling input and out in AI Agents - Multimodel in AI Agents - Adv agents input handling	-Fuzzy Output Handling pydantic in Agent -Multimodala in Agents -Adv agents input handling
Day 38	- Federated Machine Learning - Decentralized Data Explain - Federated Learning Project	-Federated Learning Project

---

Course References

• MIT introtodeeplearning

Reference Books

Designing Machine Learning Systems	Deciphering Data Architectures	Building an Event-Driven Data Mesh	Building Generative AI Services with FastAPI

Course Proejcts

- ML_in_Prod_batch_1_proj

- ML_in_Prod_batch_1_proj2

- Secure MLflow server setup for production

- Distributed training with Custom Spark Cluster

- RAG-for-NPC-Game-Characters

- Agentic RAG for Real world problem

- Text and Audio Generation in Serverless

Connect Me

If you're interested in my course, feel free to connect with me.