4 *Real* Machine Learning Projects That Get You Hired - No More Tutorials!

The video argues that completing follow-along tutorials is insufficient for landing an AI/ML job, and instead emphasizes building original, self-motivated, end-to-end portfolio projects that you scope, ship, and deploy independently. It showcases four standout examples—Shelf Scanner, a book-discovery app using GPT-4o Vision; Pack Vote, an AI travel planner with a model gateway and A/B testing; a full MLOps pipeline with ZenML and MLflow; and a humorous banana-ripeness prediction project—each demonstrating production-grade features like monitoring, rate limiting, and feedback loops. The key takeaway is to solve real problems with rapid MVPs, add engineering maturity, and manage time effectively using ACFlow to complete impressive, resume-worthy work.

• ▶ 0:00 Follow-along tutorial projects aren't enough to land an AI/ML job—you need self-motivated, end-to-end projects you scoped, built, and shipped independently.
• ▶ 0:17 Impressive projects are original and not cookie-cutter (no ChatGPT-generated ideas or common Kaggle projects) so they stand out on a resume.
• ▶ 0:50 Plan projects by identifying any problem you want to solve, doing the simplest system design (data, storage, model, deployment), and aiming for an MVP as fast as possible rather than perfecting the spec.
• ▶ 2:09 Shelf Scanner is an AI-powered book discovery app built to solve the real annoyance of not recognizing books at a sale; it lets users input preferences, photograph a shelf, and get personalized recommendations with options to save or buy.
• ▶ 3:06 The AI pipeline avoids custom training by using OpenAI’s GPT-4o Vision API to read book spines, with Google Vision API as a fallback, then GPT-4o again to generate recommendations based on reading history.
• ▶ 3:43 Production-value features were added: caching for book metadata, rate limiting to control costs, an admin dashboard for monitoring/alerting, plus tests and security best practices.
• ▶ 4:37 Weekly planning with ACFlow is key to finding time for portfolio projects: a universal inbox pulls in email, GitHub, and Notion tasks, then drag-and-drop time blocking and AI Copilot scheduling make the week visible and manageable.
• ▶ 5:52 Project #2, Pack Vote, is an AI-powered group travel planner that solves real coordination pain: it sends SMS surveys to collect preferences, generates personalized destination recommendations via AI, and uses rank-choice voting to reach consensus.
• ▶ 6:45 Pack Vote demonstrates higher AI engineering maturity through a model gateway for OpenAI/Anthropic/DeepSeek, versioned prompts with automated A/B testing, and production features like monitoring, rate limiting, and full testing.
• ▶ 8:14 The project demonstrates a real end-to-end MLOps pipeline, not just a notebook model: it includes data pipelines, automated training, validation, deployment, monitoring, and feedback loops.
• ▶ 9:04 Pipeline orchestration is built with ZenML and integrated with MLflow, creating separate training, deployment, and inference pipelines with automatic experiment tracking for reproducibility.
• ▶ 9:30 The system automates deployment decisions by checking precision/recall thresholds, then serves the model via a Streamlit app that collects user feedback to S3, closing the loop for future training.
• ▶ 10:36 Project #4 is a computer vision regression task that predicts "days until banana death" using a custom dataset of banana photos taken daily until spoilage.
• ▶ 11:15 Instead of a native mobile app, build a simple cross-platform web app with a camera button, minimal UI, and a prediction with a confidence interval.
• ▶ 12:10 The project works because it's humorous and relatable, yet demonstrates a full ML pipeline: self-collected data, fine-tuning a pre-trained model, and deployment with FastAPI.
• ▶ 12:52 The section wraps by centering on building an impressive portfolio project as the main takeaway.
• ▶ 12:56 Viewers still learning fundamentals are directed to watch the "AI Engineering Roadmap" video next.
• ▶ 13:02 The speaker thanks the audience and signs off, promising to see them in the next video.

• ▶ 0:00 Introduction & Project Planning (0:00 - 2:09) - Explains why follow-along tutorials don’t get jobs and sets up the problem-to-spec project approach.
• ▶ 2:09 Project #1: Shelf Scanner (2:09 - 4:37) - Walks through the AI-powered book discovery app, plus extra-credit and production-value features.
• ▶ 4:37 ACFlow & Project #2: Pack Vote (4:37 - 7:56) - Covers making time for portfolio projects, then the AI group travel planner with versioned prompts and A/B testing.
• ▶ 7:56 Project #3: Are You a Cat? (7:56 - 10:36) - Demonstrates a production MLOps system project focused on ML engineering beyond model building.
• ▶ 10:36 Project #4: Days to Banana Death (10:36 - 12:55) - Builds a banana spoilage predictor from data collection and regression modeling to deployment and why the project works.
• ▶ 12:55 Outro & Next Steps (12:55 - 13:06) - Closes with encouragement and recommended next steps for learning fundamentals.