A typical working day in the life of an NLP Engineer: Kaja Verhoeven-Zupanc
A typical working day in the life of an NLP Engineer
Since I became a mom, I don't need an alarm clock anymore. Waking up at 6 every morning became a habit. I usually spend my early morning with our daughter and my husband, playing along with whatever crazy idea she gets, not being bothered by my phone or any other outside distraction. We have breakfast and get ready for the day together. Usually, I would hop on my bike around 7.15 and drive to work, but since COVID-19hit the town, I get an additional 45 minutes of family fun in the morning.
I usually head to my home office around eight o’clock with my green tea. It's the most peaceful hour of my day; no kids screaming, majority of my coworkers are usually not online yet, and I can enjoy my green tea while going through my to-do list, my emails, and reading about the details of a newly proposed language model or another breakthrough in the AI world. Sometimes if I have a specific hard task to tackle, I would leave it for the morning so I can focus on it uninterruptedly and with a fresh mind.
Mornings are the most productive part of the day for me. After starting on my first task, we usually have a stand-up meeting at 9:30 with my team. In the pre-Corona world that would actually mean us standing in a circle at the office and reporting what we did the day before, plans for the day, discussing any open questions, ideas, bugs ... Now all of it happens via Teams, but I like it since it also offers some socializing that we all miss so much lately.
Afterward, I continue with my tasks for the day. At the moment of writing this, I'm doing some data analysis first. I need to get some more insight into the data of our client to decide on the approach for solving the problem at hand which could best be described as something in between information extraction, multilabel classification, and entity extraction. It turns out some preprocessing in combination with pandas-profiling gives me enough insight into the data. I decide on the strategy to obtain the first baseline results: a pre-trained language model to obtain sentence embeddings in combination with cosine similarity. I will use three different language models (Dutch specific as well as multilingual) to compare the results. I should be able to write all the code in one afternoon and get the results at the latest next morning to compare different models.
In the office, we usually head out to get some lunch around noon and then eat together in our common area or sit outside on the benches next to the MAS with the view over the little marina. That used to be quite some fun! Now I get to spend my lunchtime with my husband, exchanging some interesting facts we ran across since the morning and synchronizing the rest of the day. I'm lucky enough that he also comes from the NLP world, so he knows what I'm talking about when I say: "Did you see? In Groningen, they trained a Dutch GPT-2 model" or something similar.
After lunch, I'm having some fun with language models. I love training a model from scratch or some finetuning, nevertheless, I'm curious if the multilingual models will match the results of a Dutch-specific language model on the task at hand. Writing some scripts and running it on an Nvidia v100 could never be boring! I have a short meeting with a customer in between, a nice break from coding, and a coffee break. Time flies and to compare all the results I'll have to wait until the morning. I usually stop working between 16:30 and 17, depends if it's my turn to pick up our daughter at the daycare or not. Today it's a bit after 17 and I can already hear her screaming "MAMI" from the front door, so it's time to call it a day.
Uncharacteristically Belgian ;): it's sunny today so we'll play in the garden for a while and catch the last part of the daylight before heading in. After the dinner and her evening ritual, my daughter is asleep around 7 pm and we get the evening for ourselves.
Our evening often includes some household tasks or other grown-up obligations that I'm honestly no fan of. Sometimes some work still pops up, if a deadline is approaching or if I just have to run something overnight. However, we also try to find enough time for Netflix, reading, catching up with friends (online), some yoga in my case, or just plain talking. Don't judge, but I'm always a fan of a good romance novel so tonight I'm reading Book 4 of the Bridgerton octology! :-)