Startup Update
Week 16: It’s getting interesting
Hello, and welcome to my summary for week 16, 2023!
As you may know, I run a machine learning company in Sweden called NextML, and a few weeks ago, I committed to a challenge to share and commit to my work publically.
If you haven’t read that, you can find it here:
I like the idea of sharing my work with people who find entrepreneurship exciting and hearing others’ opinions about our challenges.
It also helps me clear my mind, and hopefully, what I learn can add value to others.
Let me know if you enjoy the story and want me to continue posting content like this, preferably by clapping, following, and commenting! :)
With that said,
Here’s my summary of week 16.
Some background about where I’m at right now
At my company NextML, our main product is an infrastructure monitoring solution called DeepInspection, explicitly designed for railways.
We collect images using cameras mounted on measurement trains and develop machine-learning algorithms to detect damage.
Here are a few examples:
Our algorithms have analyzed almost 100,000km of railway and processed over 1.5B images.
We started building the product in 2019, and it’s our company’s primary revenue source.
Since then, I’ve continuously tried to sell our product to additional infrastructure owners and railway service companies, but without success.
But for some reason, that changed in 2023. Suddenly, I receive answers and interest from close to every company I contact.
Why the change in interest?
I’m glad I asked; here’s what I think.
One theory is the boom in artificial intelligence, but I doubt that since most companies I talk to already work with deep learning and computer vision.
Instead, I believe they’ve developed their own solutions for a few years and realize that it’s not as easy as they thought.
Almost everyone comes from the railway industry, and I’ve seen many examples of suboptimal approaches to machine learning.
One company had annotated 2 million images to detect a component that doesn’t require more than 100–200 images. The idea that you must have a ton of data lives on but is not very accurate.
Another time, I talked to a machine learning team at another company that didn’t know that you should design how you sample images during training.
When you have billions of images, it doesn’t make sense to show the easy ones to the algorithm during training in the same way as more complicated examples.
Our team has a machine learning and software background, and we worked with deep learning for anomaly detection before we started focusing on railways.
It’s clear that we have an edge.
Anyway,
During 2022, we’ve continuously added new damages that our algorithms can detect, and now we’re in the lead.
When I send our presentation to potential customers, they now see that we have solved challenges that they still have, and that’s why they respond.
It’s not our only product
DeepInspection is our main priority, and the product pays our salaries. Right now, our revenue is close to $1M.
So not a lot, but some companies make less.
However, for you to make sense of this newsletter, it’s good to know that we have other products we’re working on as well.
Here’s a list:
- A Stable Diffusion API that we think is better than all open-source alternatives.
- A Photoshop plugin built on top of our API that allows users to use generative AI for photo editing.
- A Stable Diffusuin plugin for Aseprite built on top of our API
- A Joint Venture in the media production industry where we want to find valuable use cases for generative AI.
I know that split focus isn’t ideal for building a great product, and that’s been our primary challenge as a company since launch.
We certainly have a history of digging new holes instead of focusing on one thing.
It’s almost only because of me. I’m too impulsive when it comes to jumping on new ideas.
Perhaps this newsletter is also a distraction, but I feel that it helps me think clearly.
Time will tell.
That was a lengthy background, but it’s a new newsletter, and I want my readers to know what’s happening.
Let’s continue.
Here’s my current priority
So, based on what I wrote above, you don’t have to be a genius to know how I should spend my time.
Since we received a surprising amount of interest in DeepInspection, my highest priority is, of course, to contact other railway companies and sell our product.
Even though we’ve developed other valuable technology, it’s time to make the most of the opportunity right before our noses.
A single additional customer could potentially double our revenue.
But I don’t want to talk to every single company right away. We still have much to learn about making our product as valuable as possible.
Also, infrastructure owners are behemoths, and they might now want to work with a small company like ours.
What if we don’t exist a few years from now?
Long sell cycles
Another challenge is that sell cycles can last for years, and getting second chances won’t be straightforward.
It’s probably dangerous to talk to all of them immediately since customer cases are the best selling points, and we only have one good one so far.
I need to tread carefully and gather valuable information from each conversation.
One thing I’ve learned, for example, is that no one has managed to create good computer vision algorithms for damages to the railhead.
So that’s a clear priority for our machine learning engineers.
Other things I’m working on
At NextML, I’m responsible for sales and planning, but since programming and learning new things make me happy, I always have some passion projects going on as well.
Obviously, I make sure that these projects benefit the company.
When I’ve done everything I must do, I develop our user interfaces to our Stable Diffusion API.
I aim to launch the Photoshop plugin this week, but some bugs are still in the way.
Here’s the plugin if you haven’t seen it:
I’m also learning SwiftUI with the help of ChatGPT, which I’m writing about on Medium.
Here’s a link to those articles.
Accomplishments
As I started writing this post, I completely forgot that it was supposed to be a summary of last week, but finally, here we go.
The most significant accomplishment last week was that I had a great meeting with a company manufacturing and selling measurement trains to infrastructure owners.
I’ve always considered them a perfect partner for us, and I would scream with joy if we could have our algorithm running on their trains when they sell them to their customers.
The meeting was more or less an interrogation about our product and its capabilities, and it went significantly over time.
It was clear that they’ve worked a lot with computer vision but surprisingly little with deep learning.
I’ve been an entrepreneur for long enough to know that one good meeting doesn’t mean anything, but it’s still better than a bad one.
They made it crystal clear that they are interested and want to do two tests on their image data.
First, a test where we can tune our algorithm, and next, another test without tuning.
That’s perfect for us, and it tells me they understand machine learning.
Other accomplishments
I’ve quickly become a decent SwiftUI developer thanks to ChatGPT.
I’m trying to do a light-weight version of the Photoshop plugin; here’s what I have so far:
That’s about it!
Things I learned
I mostly learned more about railway damages, and that’s probably a bit too geeky for a newsletter post!
Basically, we had a four-hour-long meeting with our primary customer to talk about improvements to our product and algorithms.
It’s fantastic to have a customer that wants to help us make the product better. They are also keen on sharing contacts and answering any questions we have.
I consider us very lucky to have a collaboration like that.
Goals for next week
Next week, I have two exciting meetings with big infrastructure owners in North America, hence the post’s title.
What’s fantastic about North America is that the companies are privately owned, so we don’t have to go through public procurements.
And I think they are serious about the meeting.
Last week, I received a list of 20 questions to answer beforehand, which tells me that this is a solution they want.
I fear they think we’re too small and don’t believe we can deliver on the scale necessary. I would say that’s more likely than not.
It’s certainly a valid concern for a company responsible for a country’s critical infrastructure for the foreseeable future.
Since I’m an honest person, I’ll cross my fingers and give them the facts.
I’m confident we can deliver, and I’ll do everything in my power to at least get the opportunity to run a test on their data.
Other trials
We also have upcoming trials on data from Denmark and Norway.
These tests are more straightforward because it’s the same system that collects data as in Sweden. So basically, we know that our algorithm will perform straight away.
Hopefully, they want the complete treatment after the initial tests, but that remains to be seen.
They are not the type of customers that would double our revenue, but more real cases would make a big difference in our sales.
If I have time
I know from experience that more goals mean less success, but there are things I want to do if I have time.
Launching the Photoshop plugin is one such thing, but to do that, I need to find time for debugging.
Everything else is in place, so taking those final steps and testing the market makes sense.
I’m not expecting the best reviews, but I hope it’s good enough to get feedback and that users don’t leave within minutes.
I also want to continue developing my iOS app and learn more about Swift. It’s going great, and it’s a relaxing activity.
Hopefully, I have some cool features to share next week.
Oh, and I need to finish the main quest of Hogwarts Legacy!
Maybe that’s my primary goal.
Final words
Fantastic, you made it to the end!
Thank you so much for reading, and make sure to clap and follow if you find this content exciting.
You can also subscribe to The Startup Saga for even more updates
Thank you for your time; bye for now! :)
