A brief introduction about yourself
I am Spandan Anaokar, a final year student in Engineering Physics. I am an ML enthusiast and I like research in ML, especially Natural Language Processing. This was my first internship and it was quite enlightening.
How did you find out about this internship and what made you
pursue it?
Microsoft came at Day 1 through the Placement Office and it was the only internship which had a ML/DS role at that time. Being able to work in a top-tier corporation in my favourite field, there was no reason to refuse.
Could you describe the selection/ interview process?
Test Experience
The test was quite manageable overall. The first program required cleaning and filtering a pandas dataset using various pandas operations, which was fairly straightforward. However, the second program posed more of a challenge. It involved completing a partially written class by implementing specific functions, including inheriting a class from Scikit-learn. One particularly tricky part was creating a function to replace NaN values with the average of the remaining data, after dropping a certain fraction of extreme values. This task tested our understanding of Python’s object-oriented concepts, instance variables, and methods.
Interview 1
The interviewer, approachable from the start, focused on my NLP projects, particularly word vectorization. He delved deeper with follow-up questions after my explanation. The challenge presented was identifying English or French text. Initially, I suggested a word dictionary, but the interviewer highlighted its limitations with errors and short text. His hint about the frequency of “ze” in French led me to propose using large datasets to learn letter combination probabilities for both languages. I explained how Bayes’ theorem could determine the language and suggested n-grams for better accuracy. Finally, I wrote pseudocode based on this idea, drawing on my bigram experience. The interview ended with a discussion about the job roles.
Interview 2
The second interviewer started with introductions, then dove into linear regression with detailed follow-ups. Noticing my lack of formal ML experience, he switched to probability, specifically finding a sample distribution from a known one. While I struggled, I offered an attempt. He then moved on to voice gender identification, where I admitted limited Speech Processing experience. I mentioned audio-to-vector conversion and suggested using an RNN for training. The discussion then covered RNN backpropagation, limitations (including vanishing gradients), and solutions. We talked about handling skewed datasets through data ratio adjustments and error weighting. Finally, he asked about model speedup. When I couldn’t answer fully, he explained Singular Value Decomposition (SVD) related to PCA. The interview ended with him opening the floor for my questions.
How did you prepare for the internship?
My ML-focused preparation proved effective. DSA questions, while not typically a major hurdle for my profile, were manageable thanks to my CS213 coursework and practice on CodeChef. For ML concepts, I relied heavily on Andrew Ng’s courses and the comprehensive resource deeplearningbook.org. To solidify my understanding and coding skills, Kaggle was invaluable for practicing with real-world ML models and problems. Additionally, I explored advanced topics like RNNs, LSTMs, and Transformers, crucial for Natural Language Processing tasks.
Could you brief us about the work allotted to you?
I cannot say much about the exact work. However what happens is that interns are assigned an end to end project. You can discuss with your mentors and do the work in your own way as long as you meet the goals that the project is supposed to meet. I worked with LLMs which I had some experience with prior.
How has the experience of working been? What was the
process of shifting to a new city like?
The internship experience was quite great. I enjoyed the work culture there quite a lot. The PG I stayed at wasn’t great. There were a lot of issues and there was no food provided. I have heard some other PGs have food, but it is not great as such. Shifting is not that difficult, though it is better if you have some help there.
What were the challenges that you had to overcome during your internship? Could you share some of the learnings you picked up along the way?
One of the main challenges is that the corporate lifestyle is very different from academics. It took some time to get used to the 9-5 lifestyle. On top of that, working in a team, interacting outside of the project etc are also important. An important learning which I picked up was to not be constrained by the task. You can and should change the task itself if you find it is better. Compared to listening to profs, you do not have to achieve any tasks but rather learn to define the tasks yourself too.
Any tips for juniors for making the most out of the opportunity to do an internship?
One piece of advice would be that an internship is not just work. You will learn a lot more about how life at the office is. We should try to connect with people and get advice from the people there. Whether you continue working there or not, an internship is a great opportunity to interact with all the great minds working there.
