ML System Designs
— Lets cover different ML system design problems
Introduction
In recent years, machine learning (ML) has become a cornerstone of innovation across various sectors, from healthcare to finance to entertainment. However, as this field evolves, there remains a significant scarcity of comprehensive resources focused on effective ML system design. To address this gap, I am excited to announce the launch of a series dedicated to exploring key areas in ML system design, including large language models (LLMs), generative AI, computer vision, and recommendation systems, will start with generic ML designs and later we will dive deep into difference ML applications.
Why This Series?
The rapid advancement of ML technologies often leaves practitioners overwhelmed by the complexities of system design. Whether you’re a newcomer eager to learn or a seasoned professional looking to deepen your knowledge, having access to structured, practical resources is crucial. This series aims to provide clear guidance, best practices, and insights that anyone can apply to their ML projects.
Areas of Focus
NLP related problems
Here, we will explore NLP-related ML Design problems, covering project scoping, data pipeline, and ML pipeline. We will highlight the trade-offs of using different approaches, including baseline models, RNNs/LSTMs, and Transformer-based architectures (including LLMs and generative AI). Additionally, we will discuss offline and online evaluation, model deployment, model serving, as well as monitoring and maintenance.
- Toxic Content classifier
- Language Translation
- Chatbot (Question answering chatbot)
- Text Summarisation
- Text Tagging
Classic Computer Vision problems
Here, we will explore CV-related classic ML Design problems, covering project scoping, data pipeline, and ML pipeline. We will highlight the trade-offs of using different approaches, including baseline models, CNN, and Transformer-based architectures. Additionally, we will discuss offline and online evaluation, model deployment, model serving, as well as monitoring and maintenance.
- Object Recognition
- Object Detection
- Object Segmentation
- Object Trackking
- Face Recognition
3D Computer Vision problems
Here, we will explore 3D CV-related ML Design problems, covering project scoping, data pipeline, and ML pipeline. We will highlight the trade-offs of using different approaches, including baseline models, 3D-CNN, GANs, VAEs and Transformer-based architectures (like Diffusion Models). Additionally, we will discuss offline and online evaluation, model deployment, model serving, as well as monitoring and maintenance.
- 3D Object Reconstruction using 2D Image
- 3D Scene generation using Text descriptions
- Avatar Generation
- Problems related to Stereo Camera inputs
- Problems related to Depth maps
Recommendation Systems
Here, we will explore how to design the recommendation systems, covering project scoping, data pipeline, and ML pipeline (candidate generation or retrieval, ranking stages). We will highlight the trade-offs of using different approaches, including baseline models, Two-tower networks and complex networks. Additionally, we will discuss offline and online evaluation, model deployment, model serving, as well as monitoring and maintenance.
- Feed Ranking System
- Ads Ranking System
Multimodal related Problems
Here, we will explore how to design the systems with Multimodal capabilities, covering project scoping, data pipeline, and ML pipeline. We will highlight the trade-offs of using different approaches. Additionally, we will discuss offline and online evaluation, model deployment, model serving, as well as monitoring and maintenance.
- Multimodal Search Engine
- Multimodal Dialogue Systems
- Interactive Voice Assistant with Visual Context
Knowledge Sharing and Collaboration
The goal of this series extends beyond sharing my insights, I want to foster a collaborative community. I encourage readers to engage, share experiences, and contribute ideas. Here’s how you can be part of this journey:
- Feedback: Let me know what topics interest you the most or challenges you face.
- Case Studies: Share your experiences with ML projects — successful or otherwise — as these can provide valuable learning opportunities for all.
- Collaborative Learning: Join the discussion to help build a supportive network for exploring new ideas and solutions in ML.
Conclusion
I am thrilled to embark on this journey of exploring machine learning system design with you. By addressing the scarcity of resources and sharing knowledge across various domains, we can empower one another to build more effective and responsible ML systems. Stay tuned for the first release, Together, let’s unlock the potential of machine learning through informed and thoughtful system design.