Amygda in 2023. Running on 0 VC $

I haven't written in a while, and I mean like 'not-for-the-past-year' while. And so my thesis of building in public took a back-seat. Not that this one post is going to undone the shame of not writing for a year, but who knows, one less thing I feel guilty about!

For those who are new to Amygda AI - we speed up troubleshooting of asset maintenance for transport industries. And prevent future incidents with AI.

We are working in foundational industries like aviation and rail.

Starting life with pre-seed funding in 2021 from BrightCap Ventures and then adding to it with funding from Boeing and Tawazun Council in 2022. Most folks know us as a VC backed company.

And I am proud to have the investors we have.

But in 2023, after some initial discussions with VCs, I decided to focus on just building the business.

Building the business.

Yeah, like that's easy?

It's easy to say, let's build a business. Some of the consequential decisions were (I call them the 5 Qs I need to answer)

1. cashflow?
2. slow down AI dev?
3. how do we grow team?
4. how do we scale the business?
5. And importantly, who will pay our salaries?

The truth is, as VC backed companies know, you are investing someone else's money to build the business. And hope along the way that investment is recouped by prospects buying into the vision of the startup and become customers.

...We were never one of these startups.

In 2021, the first $ into the business was from a customer. Then came the VC $.

In 2022, the ratio of revenue to VC $ was 50:50.

In 2023, the we 100% revenue funded, $0 VC funded.

Coming back to the 5 Qs I needed to answer. First I spent some dark days, those who build businesses know exactly what those dark days look like. I don't want to dwell on them, I will have more of these.

These days are exhausting, mentally deteriorate you a lot, and can sometimes feel like everything around you is burning. Hell, you even question if you are good enough.

In the past 18 months 100s if not 1000s of business owners have had them. Startup founders a lot more.

Anyways, I wanted to answer the 5Qs. And after lots of procrastination.. the answer became clear!

1. cashflow?... sell what you do, or die
2. slow down AI dev?... no, speed up and die doing it
3. how do we grow team?...don't grow the team, until it's uncomfortable
4. how do we scale the business?... don't yet, just focus on customers right now
5. And importantly , who will pay our salaries?... those who believe in you, customers

...Is that it? That's the silver bullet?

As Ben Horowitz eloquently put it in his book The Hard Thing About Hard Things

there are no silver bullets, only lead ones”.

With the decision making out of the way. It was time to stop thinking about VCs and focus all efforts on building the business.

We halved the team, put some conversations with non-committal customers on the backseat. And put double the effort on existing customers.

Q1 of 2023 was pivotal in becoming self-aware about that path.

Moving into Q2 and rest of 2023

Focussing on building the business meant we became selective about the prospects, partners, and projects we got involved in.

Infact I learnt a lot more doing less. Because I could understand why it worked.

Product development

We don't do IF...THEN....ELSE statements in the name of AI. Like below.

def simple_AI(question):
    if question == "What is the meaning of life?":
        return "42"
    elif question == "Can you pass the Turing Test?":
        return "If I say yes, will you believe me?"
    elif question == "What's the best way to make a sandwich?":
        return "Use bread, or a tortilla if you're feeling adventurous!"
    else:
        return "Hmm, let me Google that for you."

# Let's ask our 'AI' a few questions
print(simple_AI("What is the meaning of life?"))
print(simple_AI("Can you pass the Turing Test?"))
print(simple_AI("What's the best way to make a sandwich?"))
print(simple_AI("What is the secret to happiness?"))        

As a team we actually focussed a lot more on reducing infrastructure costs, removing inefficiencies, and automating where possible.

What we did trade-off against is, that we only focussed on customer projects. We have lots of automations to build, and integrations that we know will be needed in future. But that's something where we need to put our own effort and will need to build bandwidth in 2024 and beyond.

Development of Amygda's Autonomous AI engine

Our Autonomous AI engine with its reusable and flexible AI models continued it's validation on live projects and R&D. This is powering service development for a large aero OEM and a large airport. And is being tested with a rail customer.

Utilising Machine Learning (ML) is still misunderstood by VCs. But I find more understanding and knowledge on customer side.

For example, OpenAI is nothing without its researchers and data scientists who are working manually behind the scenes and over years have built the data pipelines and models (MLOps). Whilst if I say we have to hire data scientists (we have 2 right now) VCs don't understand why - like AI is magic built out of thin air?

Customers on the other hand, ask knowledgable questions like how many labels do we need, how do you deal with data quality, and the differentiator we have - how are the models equipment agnostic? Applicable to many equipment in the transport sector!

We are winning trust based on evidence. Not marketing.

P.S. our work on flexible and reusable AI models is now downloaded at all major consultancies, universities, and leading companies in urban air mobility.

Our launch of Maintenance Copilots using Generative AI

This is our enterprise Generative AI application designed specifically for the unique challenges faced by operators and OEMs in the transport industry.

Leveraging the latest pre-trained transformer models, Amygda's Generative AI for transport application transforms maintenance and operations practices to enhance knowledge transfer and operational efficiency across the industry.

We have signed up a design customer in transport, and will be announcing two more early next year.

Working with real proprietary data (not available in public domain) with real users in foundational industries like aviation and rail.

Even though we halved the team, we focussed on a problem that the customer has which could be resolved exponentially faster with the advent of LLMs (APIs and OS LLMs both).

This is an emerging tech and where we are getting lots of interest.

The Copilot doesn't just talk - it reasons too, has agents that work with documents and dabase, running queries and letting users do tasks in natural language (since we built our own ML models, we can chat and ask for the Copilot to run say the forecasting model to show forecasts).

Leveraging LLMs to build our Copilot application is not the same as ChatGPT (which is a great benchmark for a conversational application).

P.S. we are tackling time-series data too with Gen AI working with Loughborough University we publish our work and paper on using deep generative models for aircraft data.

Eventually we believe foundational models will be multi-modal.

Commercially we proved resilient

Have I said before, I don't know how to sell?

Imagine that, as the CEO and startup founder - what an absolute mess!

After enough punches, I knew two things in sales I needed to do better:

1) Learn how to get up faster

2) Ideally... don't get punched

1 is self explanatory. It doesn't get easier, each one still hurts. But I know it will happen.

On 2 - oh this IS hard.

Not getting punched in sales meant, I had to only focus on the most likely wins. Say no to discussions I didn't think were productive. And reflect hard on ROI for each investment I was making.

As a technologist ROI is a shit-show. It's a constraint on the beautiful dream I have.

As a business, having to earn our salaries, ROI is a great filter.

In 2023,

Aviation still accounts for 80% of our business.

Rail accounts for 20% and is growing.

As a team in a few days we will be back to the same level as before halving the team.

So, without making this long, can you finish up already?

Yes, I should.

The takeaway from this post is simple, Amygda AI has continued to grow and be resilient.

We ran on $0 VC. Being customer funded through revenue is not a bad thing.

Some things went faster, some others we had to slow down on.

Raising funding from VCs isn't a bad thing either.

We will raise funding for the right thing.

Building a business is hard.

I love aviation.

I love rail.

I ?? my team.

This is possible thanks to the team, my co-founder Shaheryar (Shery) Khan , Szymon Orchowski , Asya Bodeva , and Mani Krishna M. .

And for the one person who backs me through almost everything I ask Georgi Mitov .

...Pssst what's in store for 2024?

That's for another day.

But if we haven't spoken in a while, drop me a line, you can see where I've been hiding, I need to get in touch with many more friends.

If you like what you read, share with your friends.

If you hate it, share with your enemy and annoy them too.

Happy Holidays and Ciao.