Academia.edu Innovation & Technology Culture

How is your team integrating AI and ML into the product development process, and what specific improvements have you seen as a result?

Many engineers use AI-powered coding assistants to speed up our development time. We have also used AI tools to quickly analyze and label data allowing us to perform analyses that uncover key business insights or fine-tune and deploy ML models much more rapidly than we could before.

What strategies are you employing to ensure that your systems and processes keep up with the rapid advancements in AI and ML?

At Academia.edu, we are always constantly building and learning. Engineers are encouraged to take time to explore, learn and keep up to date on the latest developments in AI/ML. 

We also have a very strong demo culture at Academia.edu where we are encouraged to take time to build small-scale demos of promising new technologies we find to validate the promise of those technologies. Many of our most exciting products and features started as small-scale demos an engineer built after learning about how a new technology could be applied at Academia.edu.

Can you share some examples of how AI/ML has directly contributed to enhancing your product line or accelerating time-to-market?

We have used AI/ML to build advanced search tools that allow researchers to search among millions of academic papers to identify papers and data. We also use AI to identify key researchers to invite to publish in our academic journals. Additionally, we use AI/ML to power reading recommendations that ensure our users can stay up to date on the latest research.

What tools support your day-to-day work?

My daily toolkit is focused. Jira for engineering tickets and for coding I've landed on Claude Code and stayed there. I work with a lot of inertia so once I find something that works I'm like, okay, I want to stick to this.

That said, I have to be very skeptical about the output. The scary part is the results that look almost right, but then there's something very subtly wrong about it. My rule is I'll never trust output until I've verified it myself. 

If I'm writing a SQL query and having AI assist me, I'll make sure that the results look like what I think they should look like. If someone else has already created a dashboard I'll make sure my query has the same results as that dashboard. It depends how high the stakes are. If you're vibe coding a productivity tool for yourself it's not really going to matter if it's slightly wrong. But if the company is making a decision based on it, it's worth putting in that extra effort to verify it. If I'm ever going to show my work off to someone else, then it's probably worth putting in that extra effort to make sure that what it actually told me is correct.

How does your team experiment?

I'd say it's semi-structured. We've done a lot of experiments where it's like, okay, for this week or two, you're going to try out this new workflow. But I say semi-structured because usually there's not a specific metric that we're tracking. It's more vibes based. Do you feel like you were more productive? Did you enjoy the experience? What things worked well? What things didn't work well? It's a more qualitative evaluation overall.

Especially this year, there's been a lot more people taking initiative and experimenting on their own. Some team members gave a presentation on their AI workflow or how they’re using AI. I even gave a presentation on my own experience. There's been a lot of people just kind of doing stuff on their own, figuring out how it works for them and then sharing it. Academia has a supportive environment to do that sort of thing. Everyone has their own unique workflow, so I ask, "Oh, what are you doing there? That's cool, I haven't seen that before." There's a good mix of styles and tools.

How does your company adapt to change?

It's tough because you have to find a good balance between doing your day-to-day work and taking time to explore and adapt your workflow. Especially today where the tools that are available change really, really quickly. You have to be a little curious.

When we first started experimenting with AI-assisted coding as a team, we were like, "Yeah, we're running this experiment, we want everyone to try this on a temporary basis. We are understanding of the fact that that might cause a slowdown in productivity." That was a good way to approach it. Everyone had a chance to experiment with the new tools and there wasn't immediate pressure to meet deadlines. There was an expectation that you were trying out new things. That protected time to experiment is really nice. 

I attended a couple of AI coding workshops and they were going over a lot of basic stuff that I felt like I already knew. “Oh wow, a lot of these people haven't even tried this before.” Whereas at Academia, we had experiments that were kind of baked into every engineer's workflow. My experience has been that my company is staying at the forefront, experimenting with new things and is willing to adapt and change.