September 17th, 2021
The Humans Behind the Bots: Aciel Eshky
At Rasa, our team is building the standard infrastructure for conversational AI. Behind the scenes, the people of Rasa come together from diverse backgrounds to solve today's most interesting challenges in NLP and dialogue management. We're pulling back the curtain to highlight a few of the humans behind the bots.
Today we're talking with Aciel Eshky, Senior Machine Learning Researcher at Rasa. We'll learn Aciel's story and explore the day-to-day projects and technologies they're passionate about.
1. Hi, Aciel. Tell us about yourself! What was your path to joining Rasa?
My path to joining Rasa is in many ways conventional, although I did take a minor detour! I started off with an undergraduate degree in computer science from King Saud University, followed by a Master's in machine learning and natural language processing, then a PhD in dialog systems, both from Edinburgh University. My PhD thesis was on modelling user behaviour in task oriented dialogs using generative probabilistic models.
When I completed my PhD, I wanted to pursue further research in machine learning, however, there were few opportunities where I lived, and I wasn't willing to relocate because I had two young children, so I jumped around between jobs to fit this constraint. I first took a short postdoc at my university, modelling athlete data from wearable sensors. After this, I transitioned to pharmaceuticals, working as a statistics software engineer at a small biostatistical consultancy for 2 years. My time in the company was interesting, and I learned a lot about randomised controlled trials and about delivering statistical models in commercial systems. However, I found myself curious about the latest development in machine learning, and in particular neural networks and deep learning, and there wasn't scope to explore this at my job. So, I decided to look for machine learning jobs, and ended up returning to academia for a 3.5 year postdoc applying machine learning to speech and language therapy. I used my postdoc to gain experience in deep learning, while also supervising MSc students working on dialog projects.
Towards the end of my contract it was time to look for a job again, when I discovered that Adam Lopez, one of our lecturers at Edinburgh University had joined Rasa in Edinburgh. This seemed like a great opportunity for me to return to dialog research and to also work with Adam. At the same time, a former colleague from Edinburgh University who was already working at Rasa reached out to me urging me to apply, so I did! Despite my detour, I felt that I had something to offer, and could transfer the skills that I learned in previous jobs to this one.
2. Take us through a typical day as a Senior Machine Learning Researcher. What types of projects do you work on?
I find that my work varies from day to day, and even from week to week, and that I need to be able to respond to things as they come up. My work generally involves prototyping new machine learning solutions, scoping new product features, or implementing new components in Rasa Open Source. I also spend a good chunk of my time working with my colleagues to improve our internal processes around hiring new candidates, developing a growth framework to progress our careers, and improving the way we collaborate with one another. I like that my work is varied and find myself reading a lot to acquire the skills I need to do my job well.
3. Which areas of your work are you most passionate about?
My main research interest is in dialog management, and particularly using off-policy and batch reinforcement learning (RL). RL directly optimises the objectives that we care about (for example, task completion) as opposed to supervised learning which mimics a dataset that only implicitly captures our objectives. RL is classically an interactive process, meaning we either need to design and build representative user simulators then use them to learn dialog policies, or otherwise learn the policy interactively in the real world, which can be risky. Off-policy and batch RL allow us to instead utilise static data for learning, much like supervised learning, which seems like an appropriate setting for commercial dialog systems. I'm excited to further explore this work at Rasa!
4. What's an important problem you're solving at Rasa?
My colleague and I are currently developing a number of methods to evaluate dialog systems. This is important because it allows us and our customers to make choices based on empirical evidence.
5. How would you describe Rasa in three words?
Flexible, empowering, and ambitious!
6. How do you collaborate with other teams at Rasa?
We currently have a research team and several engineering squads. I am originally part of research but am currently embedded in the Enable squad to work on Rasa Open Source. Our employees are spread over multiple continents and different time zones, but we collaborate effectively using a number of tools, including slack, github boards, and a wiki for more permanent content. All of our meetings are virtual and scheduled at times that are friendly to all attendees. While I'm currently part of Enable, I still attend a weekly research brainstorming meeting and still solicit feedback from my colleagues in research about things I work on.
7. What does a culture of diversity mean for you at Rasa?
Building and empowering diverse teams is a core part of Rasa's mission and I see that reflected in everything that we do. I like that my colleagues come from different parts of the world with diverse identities and experiences, and I appreciate that the company does a lot to create an inclusive environment for us and to empower us to do our jobs well regardless of our situations. For example, as a mother, I appreciate our full flexibility with working hours and the opportunity to work fully remotely, which means I'm home when my kids return from school! I've actually found that my productivity has increased due to this flexibility.
8. How has working at Rasa helped your professional development?
I learn so much everyday, and not just about the technical aspect of my job, but also about which processes work and which ones don't. My manager, Adam, involves us in all decisions, and empowers us to come up with processes that work for us. This means that we're constantly reading and learning whatever we need to in order to make that happen. We recently developed a growth framework for machine learning researchers and engineers, and now use it to assess our individual strengths and weaknesses and then decide what areas to focus on to further progress our careers. In my latest assessment, I decided to focus on delivering production software as my next area for growth.
9. What's the most interesting thing you've learned lately?
When I first encountered our continuous integration framework for model regression testing I was so impressed by how it worked. After being in academia for the majority of my career all the clever engineering behind it initially seemed like magic to me!
10. What's the best career advice you've received?
One advice I received a long time ago and often give my friends and colleagues is not to wait for a job to open up before applying, and to instead proactively reach out to people you want to work with or organisations you want to join and express your interest. I extend this idea to other areas of life and believe it's good to proactively chase what you care about instead of waiting for things to happen.
You can find Aciel on LinkedIn.
Want to team up with Rasa? We're hiring! Find open positions for our Research team and other roles on our Jobs Board.