Notes from a CTO #1: Cofounders, ChatGPT, Data Science
This newsletter was long overdue, it serves as a raw canvas of my thoughts with a small commentary on them. It is a timestamp of the things I was tinkering with this week.
Hey, it’s Bkrm from Docsumo.
You’re reading the first issue of Notes from a CTO: my raw canvas of thoughts and collection of interesting resources I found online.
This newsletter will be divided into 5 sections so that my thoughts are better organized and it is easy for you to know what to expect. My only objective with this newsletter is to provide you with immense value and create a repository of knowledge.
If this is up your alley, have a read and let me know what you think!
1. Running a startup
This section is a dump of the various facets of startups that I have experienced while running a startup. This week, let's talk about co-founders.
Choosing a partner in crime is one of the most difficult decisions to make when you are embarking on the long and arduous journey of starting a company. This will be the person you will spend at least 60-70 minutes per day so you must look out for certain chemistry between the two of you. A lot has been said about this topic, but it was initially made famous by Y-combinator.
And today, I am just trying to add my spin to it.
For me, the most important factor is trust. Your co-founder is doing their best and wants the same outcome as you, i.e., the success of our startup. We are human beings, and we think differently. 6 can be 9 and both can be right in their way.
You will have long discussions ( you will not know when these discussions become arguments), and there will be a time amidst this conversation when you want to punch a hole in your laptop because nothing is going as planned. Honestly, it is okay and part of the process because building and maintaining trust requires a lot of hard work.
Here is my framework for that.
Don’t take things personally, the next person is trying to help.
Have an open mind. You might think you are right and the next person is wrong. Even in this case make a hypothesis, make it time and resource-bound, and experiment. You will learn a lot in the process.
Talk about difficult things. But have time and place for this, it should not be in the middle of a discussion. Talking solves a lot of issues.
Appreciate when another person is right and you are wrong and never point fingers. There is no place for I told you so. (Just remember, they were wrong based on the data points that they had at that point in time.)
When the argument is over, cool down and get on a call and talk about things outside of work. You will find the true reason for their behavior.
Understand the next person is also human and has a life outside of work. You have zero knowledge of what has already happened from the time he/she woke up.
When the next person is feeling low, show them how far we have come. Look back at your team, change log, customers, and the fun you have had. Sometimes we forget to be happy with small achievements when we are chasing a larger goal.
Take ownership when you screw up. This will instill more trust! The best employees I have or had are those who told me “Sorry I screwed up, but this will not happen next time and this is how I will solve it in the future”. The secret sauce is No blame and excuse game. This is something I used to tell at Daraz, when the department used to point figures.
Is this easy? No.
Do I always follow these points and not lose my temper? Hell no.
Do I think about this point once I cool down? Hell yes.
At the end of the day, we have to optimize for our happiness.
I was discussing this framework with my friends and we thought, “Can’t we use this framework for our life partner? ”
Our conclusion was partially yes!
Honestly, you can apply this with your life partner/ anyone you have arguments with. But with your life partner the goal is not clear or simple like in the case of a co-founder (make startup successful), as there are so many external variables.
It true that end goal for a couple is to be happy, but it easier said than done.
A great article on this topic:
Together Blog on Co-founders, partners, friends, and family
2. Technology
This section is a dump of the various technologies that I have been exploring. This week, let's talk about Generative AI.
Generative AI is the next buzzword in tech. Most pundits are making predictions and some have even stated - It is the end of Google, Schools, and whatnot.
If you cut through all this noise, many industries will be disrupted due to this technology.
One of the reasons, we have not dived deep into document QA models (like https://huggingface.co/docs/transformers/model_doc/donut - the OCR free part is really good) because these kinds of models have use cases, but we are still unsure which ones are real painkillers for users and not just vitamins.
ChatGPT usage is ideal for use cases such as email writing ( I have used it to write emails more than 5 times in the last week) or where the users do not require high levels of accuracy. There are few unicorns like jasper.ai in this space. I am a huge fan of Jasper and how they have cut through the competition; more on that in the future.
Though this space is very competitive, I am really excited about Notion AI. Their usage case is on point.
A big moat in this space will come from figuring out usage cases where the incumbent players can have proprietary data and also give a highly accurate result.
Reading resources about this space :
- Sequoia Blog on Generative AI: A Creative New World
- Base 10Vc: Generative-ai-mission-critical
- CB Insights: The 250 companies driving generative AI forward
3. Podcasts
This section is a dump of the various podcasts or videos that I have been watching. This week, let's watch the Python Godfather in action.
No words can do justice to this podcast.
Strap in and brace yourself for this long but worth-watching conversation.
4. Interesting links
This section is a dump of resources that I found interesting.
Repos:
transformers-interpret: Interpretation is the first step to making a better model and it can teach you so many things in the process.
katana: Web Scraping has always fascinated me. Maybe because I like collecting data. I might not do anything with it but the ownership of data always excites me. We have about 10,000 annual statements from the last 2 years that I had scraped on one boring weekend. We just used that data this week. Had to hire a new employee to work directly with me to use it. 😂
Ntfy: Push notification for mobile and desktop: I see a lot of usage cases for this. Will I use it anytime soon? Most likely noo.
Articles:
Pray to the god of Valuation: Rushabh and I did not celebrate much when we got our $3.5 Million funding because it was a small milestone. It was someone else’s money given to us so we can make better use of it. On the other hand, we feel really elated when the sales team sends a $10K deal on our sales slack channel.
An important point to note - Today, our company shares are just paper money. When we give ESOPs to employees, the most important thing they need to remember is, it is paper. It will only become valuable when we get more customers and have growth.
All the buzz on fundraising has really screwed people's perception of valuation.
Just say No: The ability to say no is underrated. I really need to learn to say no to most things. Hopefully, I will learn this in 2023. (Maybe a good new year resolution?)
Goodbye Data Science: We started our team with just one data scientist. Now that we have grown and have to deliver and maintain APIs, the team has an equal (50%) split between Python developers and ML engineers.
I have been lucky with people, they have done whatever was required since Docsumo’s inception. From the first group of data scientists we hired, we had people taking ownership of the backend, customer success, writing rules, and solving problems using computer vision. That is how we built Docsumo.
But keeping people happy who just want to pursue data science is next to impossible. If someone is not a good software engineer before they become a data scientist, don’t hire them for your startup. Real life is not Kaggle, and 90% of the time will go into data collection and preprocessing. It’s impossible to always work on a sexy new algorithm.
I have had a tough time, making data scientists understand that it’s ok if their 3-month long work on a project did not go past the POC due to lack of accuracy. Well, good luck explaining this. ( That is the reason our job posts will only say we are hiring ML Engineers and not Data Scientists. You may ask what is the difference? A topic for another day maybe.)
You only need a few hard-core ML researchers on the team, other people can be generalists. You can make a good data scientist from a good developer but the other way around is very difficult.
Kubernetes Problems: I attribute a large part of Docsumo’s success to Kubernetes. After many experiments on Ec2, Docker, Docker-compose, and Docker swarm, K8s turned out to be lifesaver on so many fronts. It is the reason we don’t have a dedicated DevOps team until now, even with a team of 30+ engineers, 3+ k8s Custer, and more than 200+ APIs. Though, K8s can be tricky sometimes so it really feels good to have that Eureka moment.
5. Quotes/ Books
This section is a dump of the various books and quotes I have been reading.
Book recommendation: The Mom Test
In hindsight, I feel as an engineer this is first thing I should have read before starting my engineering journey“The reason bird can sleep in branch without worrying about branch breaking is, it has more trust on his wing than branch”
This quote has been my philosophy since my first job and even when Rushabh and I were thinking about starting Docsumo. A very practical way to think about this is by pondering over the worst-case scenario. This exercise helps both of us realign our focus. If you can tackle the worst case, you have achieved freedom.
A few examples to put this into perspective: what happens if this feature gets zero users? what happens if we can’t raise money? what if we have zero customers for the rest of the year? what if we have to take the horse to the barn and shoot it?
Our Slack channels have a great collection of memes.
P.S. Docsumo readers, I am biased what can I say, repurposing a meme that I only posted 😂
That’s it for this edition, I hope you find it useful.
Have an amazing year-end and I hope you get enough downtime to start the grind again in January.
See you in two weeks!
Best,
Bikram Dahal
P.S. If you learned something new today, please share “Notes from a CTO” with your friends and spread the love. ✌🏻
First Section is really helpful to be success in life, other tech stuffs looks great.
Wow. It's really helpful. Looking forward to more.