Blend

20 Sep 2018 |

I just finished working for a year at Blend, and now I’m back at UCLA as a student.

Working full-time was actually quite a bit different from what I expected. I wanted to talk a bit about the things I did, and what I liked/didn’t like about working full-time at Blend.

NOTE: Take what I say with a huge grain of salt, as this is just my personal experience (\(n=1\)).

What I Did:

At Blend, I deployed and maintained several internal data servies:

  • Airflow - Airbnb tool responsible for scheduling Blend’s ETL jobs
  • DSTB (data science toolbox) - Hosted jupyter notebooks with easy connection to our Redshift cluster.
  • \(\log R\) - Blend’s log archive pipeline, which has now ingests a variety of different data sources, such as frontend event data.

I also spent a lot of time working on a variety of more nebulous machine learning projects to improve Blend’s product:

  • using an RNN to predict user submission.
  • Trying to cluster our NPS comments.
  • using deep learning to identify “frozen” NPS comments.

Most of these fizzled out, but the frozen NPS classifier is still being used today.

I also spent a lot of time writing documentation/specs/tests, resolving on-call issues, reviewing code, and sitting in meetings.

Something that suprised me was just the sheer volume of “other” work that needs to be done, and the relative importance of it. I think the most effective engineers don’t necessarily spend a lot of their time writing code, but rather managing/organizing. This is definitely something that I didn’t anticipate doing (or at least that I would spend the majority of my time doing these things vs. writing code).

I also came out of this with a huge newfound appreciation for good PMs, they really have a huge effect on engineering productivity.

Things I like about working full-time:

Industry felt a lot faster than academia.

There was a lot of energy at Blend, which was fun. In my time there, they expanded from ~170 employees to over 300, and had a national roll-out with Wells Fargo and US Bank. Work definitely felt higher-impact than school.

One of the things I really enjoyed about working was that so long as you solve the problem, how you solved it doesn’t matter too much. You’re encouraged to not reinvent the wheel, to call into libraries, or even just buy a service instead of building in-house.

I was also given a lot of freedom in my approach, which I really liked.

Overall, I’m a lot less of a perfectionist now - I feel that I am more methodical in how I approach projects - creating a feedback look and iterating quickly.

Iteration is also inherently way more scalable than doing everything in one go (easier to explain to other people).

I think the biggest thing that I enjoy about working is simply building things and helping people.

Things I don’t like about working full-time:

One of the bad things about “moving fast” or working at I think any sort of start-up is that there’s going to be a sizeable amount of tech-debt. This was fine, but sometimes new features would get prioritized over tech-debt, which meant a lot of time working with sub-par code. Also there was a tendency to monkey-patch the system and not solve the root cause of the issue, which lead to a lot of reoccuring problems.

Of course everyone has problems with tech-debt, but I think startups in general tend to have more.

I think the biggest thing I don’t like about working is that I think you may get pidgeonholed. Any task you do once is something you’ll probably have to do millions of time. Furthermore, sometimes hard to branch out because you still have an obligation to maintain legacy code.

One of the biggest reasons that I left was that I felt that I was stagnating and I found myself burnt-out and not very productive However, I was working on a personal project, that I really liked so I knew it wasn’t so much that I was fed up with computer science so much as I was fed up with work.

One thing that was kind of frustrating was that there wasn’t much machine learning at Blend. I would say that’s a function of the age of the company - can’t do data science when you have no data. This is also tied heavily into the core product of the buisness as well. In particular, machine learning at Blend is not the product, it would just be used to enhance the product. As such, the value proposition of ML is pretty minimal given how much complexity it adds.

I think in general though, “machine learning” everywhere (and especially at start-ups) is usually much less “making cool models” and much more “fixing some broken ETL job”.

Overall though I really enjoyed the work I was doing at Blend. Even the ‘boring’ bits were still fun, and the team/work culture was top notch.

Written on September 20, 2018