What a Data Science tech interview in a big 4 looks like 😱

Last month I went through a fairly complex interview process with a big consultancy company. 💼

The position was for a Senior Data Scientist, and it involved 4 rounds of interviews, of which 3 were technical interviews. Here are all the questions they asked me.

🧠Machine Learning
- Explain what is RAG and when it's used
- What is K-fold cross validation? 
- How do you fine-tune a LLM?
- How to avoid pitfalls when training on a biased dataset?
- When to use fine-tuning Vs Vanilla LLM output?
- What is the Loss used in LLM models? 
- How to deal with high throughput in Depp learning models?
- Loss used in Logistic Regression vs Linear Regression
- How to monitor the training parameters of a NN and which tools to use?

👨‍💻Python
- What is the difference between list and tuples? 
- What are some ways to ensure test coverage in a git repository?
- How to ensure output types of a model are preserved?
- What is a decorator?

These questions are also quite relevant for many other job positions and I think they might help you in case you are applying to one of this 6-months to 1-year contract positions.
Hope they can be useful to you too!

📱 How to write your Upwork proposals

✍️ The way you write proposals is very important on Upwork. Impress your client and they will be happy to hire you.
My first proposals where so boring that nobody wanted to read them, but then I started to think like a client and developed a very different strategy.
Now my proposals on Upwork get me an interview 50% of the time and 20% of the time I get the job. This is what I do differently.

🎯 Be direct
Stop showing off your credentials, how long you have been in the industry, who you worked for. Nobody cares. What clients care about is that you are able to do what they ask for. So I just start with something like "I have done this before", "I did a similar gig before" or "I already made a sample of the work". This catches the attention immediately and if you are able to catch the attention in the first lines the client will keep reading the rest of the proposal.

🤝 Be personal
The clients are focused on their business and want to feel that you have invested a bit of time in understanding what they are up to. I address them by name (I find it in their previous reviews), make a nice comment on their business, and generally try to show that I am excited to work with them. Also, if I find out that the client is Italian but the job description is in English, I write them in Italian. This goes a long way establishing trust and makes you more likeable.

🔗 Send links to previous work
If you have done something similar in the past the chances that the client hires you increase a lot. Make it clear that you know what you are doing and drop examples of previous work that you have successfully completed. The proposals where I insert one or more links are a lot more successful than the ones where i don't put any. People are curious and will click on the links. So add your Github project, your app, a link to your published book, blog post or article. I promise the clients will click on it and spend more time on your proposal.

👨‍💻 Show skill and expertise
If the work is fairly technical you must show off your technical expertise. Drop names of libraries you would use, relevant cloud services you have used in the past and in general give the client the impression that you know the tools. If the client is non-technical (most of the time) they will surely be impressed. Bonus points if you can already make an MVP of what they want to do. It will make you stand out from the crowd.

I don't think most freelancers put enough effort in writing their proposals, and many of them even use AI to write hundreds of identical cheap messages.
Don't do that, spend some time crafting a good message and I can guarantee you that you will be ahead of 90% of the competition.

What do you think? Do you have other tips to improve your Upwork proposals or cold emails?
Would love to hear those!

Sorry guys/gals 🥺

I haven't posted in a while, super busy with many clients and traveling around Europe 🚂

But I want to still point you to a great article by Dominique, who spent the entire 2024 as a Machine Learning freelancer and wrote a very interesting review of lessons learned earlier this year. Check it out 📕

https://thisiscrispin.substack.com/p/a-review-after-one-year-of-freelancing

👋 I got in touch with Dominique on X 🔗https://x.com/DominiqueCAPaul
I asked him a few more questions about his experience as a Machine Learning freelancer
Amazing conversation, I learnt a lot and I am sure you will learn a few things too.
This is the full transcript. Enjoy! 📖


❓How long did it take you to land your first client and what channels did you explore? Mostly personal connections and word of mouth?
🔈I found my first client during the last months of my master's thesis. I was working from an AI coworking space and met the founder of a legaltech start-up who needed help building sales demos for what AI could do for the companies.

❓Have you ever used Upwork or other platforms? If yes, what do you think of them?
🔈No, never. I do not recommend them. You are being priced as a commodity and will never be able to charge more than 60€/hour. You want to make trust your main selling point. People will never go to upwork to get a critical task done where they'll be fucked if its messed up. They want someone whom they can trust to get the job done. I wrote on this in my blogpost.

❓I quote you "the ‘right’ amount of demand is one that far exceeds your capacity. A constant stream of requests gives you the confidence to raise your price. If you only get a request every two months, you’ll be afraid of not having any income". Totally agree with this. In practice, how to evaluate demand? Do you look at the job market on LinkedIn or other indicators?
🔈This is literally the number of project requests you get for actual projects. You want more people asking you to do a job for them than you have time. An indicator can be first calls with potential clients.

❓You seem to work on high rate projects. In my experience those are not so common and most clients would be scared of those. How to get there? Do you need credibility? Or is it a matter of working with the right clients?
🔈Offer a discount for the first leg of the project. Once people get used to working with you and see what you can deliver they will be happy to pay more instead of a) taking the risk of finding someone new and b) onboarding them to the companies work / the project. If you sell your work for e.g. 60€/hour and then ask for 120€ for the second part of the project you look like an asshole. If you say that you'll offer a discount of 50% for the first two weeks then the default is that it will be 120€ after that. By then the "default" is also that you're working for them and they need to go through much more effort to find someone else.

❓What does closing a client looks like for you? How long does it take from first contact to signing the contract?
🔈I have one get-to-know call where we discuss the project high-level and where i tell them how to fix the problem without me (this creates trust) and if its clear that they'd need me then I tell them how to go about it. Ofter there will be a second call where I do some basic homework before on how to build what they need, structure it into a minimal version that we can build in a week or two, and a cost estimate. Often that's enough. Those calls might be one week apart and then work might being 1-2 weeks later. My sample size is 6 projects though just to be clear.

❓You mention "It happened twice on the same project that I agreed to a fixed price only to work more than double the expected hours". How you deal with the costs of a project getting higher, or maybe missing a deadline with a client? Any tips?
🔈Yes, take the estimate for how long you think you need and multiply it by 1.5x or 2x. Or just don't do fixed prices. Q: Where do you find contractors? Platforms or personal. connections? A: Ideally friends of mine but for basic tasks I use upwork.

❓I noticed you spend for clients dinner/travel/Wework. Did these pay off? Which one gave the best result?
🔈Wework was important because I have a office wherever I go and I was working from different places. I am much more productive at a desk than working from a cafe so this paid off 100%. Meeting a client in person is invaluable. BCG and Mckinsey also spend 20% of revenue on travel because of this reason. Meeting in person builds trust and trust is the main thing that you can charge a premium for in your hourly wage.

❓You mention that one of the challenges of freelancing is that "Cash flow is very uneven" and you were in overdraft for a the first half of 2024. How do you deal mentally with it? Any practical tips?
🔈Tell your clients that you send them the bill for each month in the first 10 days of next month and ask for payment within 14 days. Most people will understand this. Sometimes stress is also a good motivator :p

❓In the future you write "I’ve decided to focus 100% on freelancing until I have enough money to sustain myself for 6-9 months, and then 100% focus on a product". This is polarizing in the builders community on X. Why you think you can't build on the side?
🔈I actually don’t think it’s polarising—it’s just a tradeoff. You can only have one true priority at a time. Imagine you’re aiming to launch a product by a certain date, and a few days before, your client asks you to build a feature by next week. You can’t do both well. If you prioritise your product, you let the client down. If you prioritise the client, you lose momentum and compromise your own goals. To build something great, you need uninterrupted focus. Momentum compounds, and the returns on full commitment are asymmetrical - being twice as good doesn’t just get you twice the outcome, it can be 10x more impactful. At the end of the day, who’s more compelling: the person who’s #2 at two things, or the one who’s world-class at one?

❓What are your plans for the future?
🔈Building general purpose machine learning models for robots to make Europe lead in Robotics and not just follow the US or China. If you know somebody working at a company using robotics or interested in automating more with robots I'd love to speak to them.
Read more here
🔗 https://open.substack.com/pub/thisiscrispin/p/robots-and-runway


If you want to follow Dominique here are a few links
🔗Substack https://thisiscrispin.substack.com/
🔗Blog https://www.thisiscrispin.com/
🔗LinkedIn https://www.linkedin.com/in/dominique-paul/
🔗X/Twitter https://x.com/DominiqueCAPaul

👋I just published my latest blog post on my side projects results for May 2025

With the full revenue and costs breakdown.
🔗https://www.tropianhs.com/diary/2025/06/06/may-report.html

👋Hi everyone I am back! 👋

This month I wanna talk about about the type of freelancing I do. I used to write these reports more often in 2023/2024, in my first year of freelancing. You can read a few of them on my blog.
When I started freelancing again at the beginning of this year I stopped updating on the financial side of it, and limited myself to say a few words about the type of work I was doing. The main reason is that I am not comfortable sharing online the numbers, since some of my clients also read my blog posts.
This is a safer space in that sense, and I feel free to share more than I would be able to share in my personal blog.

💵 To sum up In July I freelanced for two clients. I worked less than 3 days per week, and I made roughly $4200 before taxes. Which is $3100 after taxes in Italy.
Below the details.

Client 1️⃣
I am working with them since January. I won't name them , but it's a Product agency that helps B2B companies build their online products with a focus on SaaS.
The founder is a guy I met in Milan in 2018 and, even if we both moved to different cities, we still kept in touch on X. When he heard I was looking to go back to freelancing, he contacted me and we started to work together.
They onboarded me to look at their clients Analytics, take care of A/B tests and deep dive into their clients data and find insights to build a better product.
During these months I was busy setting up some metrics on GA4, looking into their clients custom dashboards and integrating the GA4 data with the custom dashboards one to evaluate the impact of our UX/UI changes on the conversion rate and the sales funnel steps in general.
Not a typical Data Scientist job, but it gives me the opportunity to be very close to the clients product, and I love working on improving products with data.
My rate with them is $48/hour and I work at most 80 hours per month. So this keeps me busy around 2.5 days per week. But there were a few months when I worked much less.

Client 2️⃣
I am working with them since a couple of months. They are based in the UK and they are friends. They contacted me since they are scaling their business and they need someone to help with the tech side of it. They tried to hire on Upwork, but got so disappointed with the quality of work and lack of professionalism, that they asked me.
My work for now is basically writing scripts to modify they Google Sheets they share with their clients. Nothing fancy and 100% Javascript code. I am not a fan of JS but I was able to do a lot of work, mainly by letting the I write the bulk of the code and fixing here and there when needed. Amazing stuff.
Since they are friends, I am not asking more than $45/hour, and they don't have a big monthly budget. I work for them between 9-10 hours per month but it might increase in the next months.

And that's it for now.
Let me know if you like these reports, I will try to make them each month, although they might become boring, since I do not plan to onboard more clients in the next months.
I have worked with other clients during 2025, maybe I will talk about them in the next posts.
👋 Bye!