to break into machine learning, then doing cookie-cutter projects and following basic tutorials is like trying to win a Formula 1 race in a go-kart.
You’ll move, but you won’t compete, and certainly won’t win.
I’ve reviewed hundreds of ML portfolios and interviewed dozens of candidates for real data science and ML roles, and I can tell you this: the people who get hired build projects that go beyond tutorials.
So, in this article, I’ll break down the exact types of projects and frameworks that actually land interviews and job offers.
They’re not easy.
But that’s precisely why they work.
Reimplement a research paper
Think about it.
A machine learning research paper is the culmination of several months of work by some of the leading practitioners in the field, summari…
to break into machine learning, then doing cookie-cutter projects and following basic tutorials is like trying to win a Formula 1 race in a go-kart.
You’ll move, but you won’t compete, and certainly won’t win.
I’ve reviewed hundreds of ML portfolios and interviewed dozens of candidates for real data science and ML roles, and I can tell you this: the people who get hired build projects that go beyond tutorials.
So, in this article, I’ll break down the exact types of projects and frameworks that actually land interviews and job offers.
They’re not easy.
But that’s precisely why they work.
Reimplement a research paper
Think about it.
A machine learning research paper is the culmination of several months of work by some of the leading practitioners in the field, summarised in a few pages of text.
The amount of knowledge in these papers is tremendous.
So, if you break down, dissect, and re-implement these papers on your own, imagine how much you would learn.
It’s kind of like trying to rebuild a Formula 1 car from blueprints — you might not have the same tools as the original engineers, but by understanding every nut and bolt, you learn how the whole machine works. And when you finally get your own version running, you’ll understand racing at a level most people never reach.
Re-implementing a paper will teach you so many skills:
- Being able to understand complex maths associated with cutting-edge models.
- Being able to implement sophisticated models using code from scratch or simple libraries.
- Being able to think creatively and apply your own knowledge to new ideas.
And the significant part is that the majority, and I mean nearly 99%, of candidates are not doing this, so you will instantly stand out.
However, it’s not easy, and I can tell you that from first-hand experience. But easy is not going to get you hired nowadays.
Now, how you go about implementing the paper could be a whole post in itself, but let me run you through the key steps:
- Read the paper. Then, reread it, and again, and again, until you fully understand what the paper was trying to solve, the algorithm used, the data, and why the results were significant and whether they are shocking or expected. Depending on your experience, this may take awhile.
- If you don’t understand certain concepts, go and learn them. This is not a waste of time, as you are actively closing the knowledge gaps you have.
- Sketch/code the high-level architecture, like the inputs and outputs, the rough design of the overall system and the structure of the ML model.
- Start implementing the simplest part and get it working.
- Build a rough working prototype.
- Optimise and try to replicate the results.
Some papers I recommend implementing:
- Attention Is All You Need
- An Image is Worth 16×16 Words: Transformers for Image Recognition at Scale
- Language Models are Few-Shot Learners
- LoRA: Low-Rank Adaptation of Large Language Models
These are mainly within the deep learning space, but you can find papers relevant to the field you want to study.
Some useful websites to find papers:
Solve your own problem
“What projects should I build”?
This is the second most common question I get asked, the first being how I got so handsome!
The thing is, most people don’t understand that the question is the wrong one to ask (the project one, not the handsome question).
If I gave you an exact project to do, there would be no story behind it in the interview.
What are you going to say?
“Oh some guy from the internet said I should build it”
Not exactly a great scenario to be in.
A project that will stand out is deeply personal to you, and you are motivated to solve it. That is much better and interesting, and it will show during an interview.
Example project
Let me give you an example of a great project.
I mentioned this story in a previous post, but I am going to repeat it to really emphasise the type of projects you should build.
At my previous company, we were hiring for a junior data scientist to work on operations research problems.
The candidate we ended up hiring had a standout project that was directly relevant to the job and was a problem they were interested in solving.
They had an interest in fantasy football (NFL) and designed their own optimisation algorithm to better allocate their player selections each week.
They even went further, reading journal papers on others’ solutions and implementing some of the ideas. See the link with research papers!
My framework
Here’s a simple framework for you to follow to come up with a similar project as the one I just mentioned.
- List at least five things you’re interested in outside of work.
- For each topic, write down five questions you would be interested in answering or solving. So, in total, you will have 25 potential questions.
- Now, think about how machine learning could help answer those questions. Don’t worry if the question seems completely impossible; be creative. However, obviously, don’t try to create robot dogs or something!
- Finally, pick one question that excites you the most. Ideally, choose something that feels just slightly out of your reach; that way, you will really learn and push yourself out of your comfort zone.
This exercise will take you 10 minutes, so you have no excuse not to do it, and will give you a project idea that will help you land a job.
Building complexity and scale
However, the idea on its own won’t necessarily be sufficient. For that, the project needs some complexity and scale.
This can be shown and expressed in different ways.
- You can deploy the project end-to-end using production code, cloud systems like AWS and containerising the algorithm using Docker and Kubernetes.
- You can use a really complex, state-of-the-art algorithm or framework. Reading research papers is excellent for this!
- You can make it so users can interact with the project, like an online application.
- You can make it solve a variety of problems, like a suite of models.
There are many options, and it’s easy to get overwhelmed.
Start and learn as you go. That’s all you need to do.
Other ideas
If, for some reason, you don’t fancy doing the above two, even though they will actually get you hired, here is a list of further project ideas.
- Ask AI for a project; give it a suitable prompt, of course.
- Enter a Kaggle competition, but you need to place well for it to stand out.
- Use an AI/foundational model to solve a personal problem.
- Code machine learning algorithms from scratch using basic Numpy, or even better, native Python only.
Now, if you want me to further handhold you, this is a list of more granular projects to try:
- Reinforcement learning for Pac-Man or any other game.
- Building a language model from scratch.
- Computer vision model for classifying images of literally anything.
- Sentiment analysis on a social media platform about a particular topic.
- Recommendation system for an App you like.
- Fine-tuning an LLM for a particular use case.
Again, I am giving high-level ideas because these need to be personal to you for them to really stand out.
After you’ve built these projects, you’re ready to start applying for jobs!
But to actually land interviews, you’ll need a rock-solid resume.
So what makes the difference between a resume that gets ignored and one that gets noticed?
Find out in my previous post below.
Another thing!
I offer 1:1 coaching calls where we can chat about whatever you need — whether it’s projects, career advice, or just figuring out your next step. I’m here to help you move forward!