I had a difficult relationship with side projects in the past. Basically I could never get anywhere with them. Thanks to a set of techniques that I called “slow programming” I started to develop side projects in a way that brings me joy. And this is the story that I’d like to share with you.
In the past when I was starting a new side project I’d try to fulfill these three goals:
- build something meaningful that people can benefit from
- use a cool new technology that I wanted to learn
- make each commit perfect
As you may suspect, these three goals are really hard to fulfill all at the same time. I’d usually give up at the stage of learning the new technology or setting up the test environment. I didn’t have any visible results at the beginning so I was not motivated to continue with the project. And a similar pattern would repeat with many different ideas.
This changed when a good friend of mine came to me with an idea for an application that he needed. He was searching for a new apartment in Berlin and noticed that for state-owned companies (landeseigene Wohnungsbaugesellschaften) offers were disappearing very quickly after being published on the website. Often it was a matter of 30-60 minutes. My friend would receive an email notification about 2-3 hours afterwards, but at this point most of the offers were already gone.
His idea was to develop an application that continuously checks the websites of these companies and sends a notification as soon as a new offer is available. And he needed my help.
We made an appointment for a “hacking evening”. I was hoping that we can get something up and running within a few hours. All in all, the functionality was simple web scraping. I wanted to use some existing solution as much as possible. I didn’t want to reinvent the wheel and I didn’t have that much motivation.
After a quick research I picked Scrapy - a scraping framework in Python, supported by a commercial platform where the code can run. I started putting up some code together as fast as possible. Soon I realised that the persistence is an important topic, because we want to somehow know which apartment offers are new and which are not, in order to avoid sending the same notifications over and over again. I tried to get the code running on the provided infrastructure, with each passing minute getting more frustrated that it doesn’t work yet, and more disconnected from the reality.
While I was getting deeper and deeper into the programming zone, my friend was
sitting next to me, getting more and more bored, because at some point I
stopped explaining what I’m currently doing. Not exactly the “hacking evening”
anyone would wish for. After my frustration reached a certain level and my energy
dropped to zero I did
rm -rf the-project and told my friend: “Well, I guess
that I need to think about it more in the next days”.
New approach - take it slow
The result of this thinking were a few guiding ideas that I decided on:
- use “boring” technologies that I already know - Ruby, Nokogiri, Rails, and Heroku.
- no more than one hour of work on the project per day
- focus on delivering the smallest possible thing that actually provides some value to my friend
- write the first simple solution that works instead of striving for technical perfection
I thought about them as defining my new mindset of “slow programming”. And this is how Neue Wohnung was born. If you look at the commits in the reverse order you can see that I started with the implementation, added tests soon, but not immediately, and configured CI after I had a running version in production.
I was working at a slow, steady pace, trying to enjoy every moment of it. As I removed the pressure of time from myself, I could actually enjoy the little details like a new Rubocop rule that I didn’t know before or read up on libraries that I was considering using.
After a few days of working I had the first version and could actually start sending notifications to my friend. This meant that I was receiving feedback early and could adjust the application based on it, without making too many false assumptions.
Was the fast “hacking evening” with Python a wasted time? No, not really. Thanks to this session I learned that Telegram is a messaging app for which it’s extremely easy to send messages programmatically and how to create my own Telegram bot. From the time perspective I think about it as a prototyping session, although during the actual session I didn’t have this prototyping mindset.
Next boring steps
The scraper was working in production (I quickly set up Heroku) and my friend was receiving new apartment offers every day. The next step was very boring - add support for more companies. It was a tedious and repetitive activity to write this code, but I was never doing it for more than an hour per day and at the same time immediately getting the business value out of it - more and more apartment offers were reaching my friend - and this kept me motivated.
I always started by copy pasting the previous scraper and the tests for it. The code was repetitive but I avoided reaching for any early abstractions. The classes were simple and could fit in a single screen. The actual business logic - getting information from the HTML using Nokogiri and regular expressions
- was unique to each company website; with very few common elements.
This approach of not making too many assumptions paid off as soon as I had to parse JSON instead of HTML. That happened when I found the first website loading data dynamically from an unsecured JSON API :) Because of the lack of any authentication it was just easier to access the JSON API directly. My abstraction was lightweight so I didn’t have to modify it. I just wrote this one particular scraper differently.
At some point I introduced the Apartment model to the application. At this
stage I wasn’t yet sure what data I could actually get in a consistent way. The
only thing that I knew for sure was that I needed to store a unique identifier for
each apartment to avoid sending duplicated notifications after each subsequent
check. That’s why I added a mandatory
external_id field. Everything else ended
up in a JSONB
properties field. This was simple and allowed me to move fast
with delivering value while moving slow with technical assumptions.
Starting to use the app myself
A month or two had passed and guess what? I found myself in a need for a new apartment. Yes, I had to move out from my flat share and find a new apartment for myself. The application that I originally developed for my friend became a useful tool for my own apartment search.
I was particularly interested in scoring an apartment in one of the housing associations (Wohnungsgenossenschaften), because they provide stable and secure renting possibilities and I believe in this organisational form. The domain situation here was different. The offers didn’t appear often, sometimes you could wait for the whole month for one housing association to publish something. However, there were about 30-40 housing associations that I was interested in.
I started with a long list of links and every day was going through all of them, mostly to find out that no new offers appeared. This was a tedious and frustrating process. So I began migrating the sources to be automatically handled by my app, one by one. From the technical point of view they were no different than the previous sources, so I didn’t have to change anything in my assumptions. Again, I was working step by step, no more than one hour per day. Each migrated source meant one link less to click every morning, the list was shrinking, and the progress was visible to me. This kept me motivated.
One day I received a notification from a housing cooperative that was particularly appealing to me. The apartment was in a neighbourhood close to my old apartment and a really good deal for the money. I immediately applied for it and from that moment on things moved fast. Next day I had a viewing and within a week I signed the contract.
The application that I developed as a side project scored me a new apartment! That means it’s my most successful side project so far.
I attribute this success to the philosophy of “slow programming” that I followed:
- put a hard limit on the time per day that you invest into the side project
- move in small steps, don’t rush, and appreciate the little learnings that come every day
- stay motivated by making a tiny but visible improvement with each step
Happy slow coding!