Hiring for Startups: Takeaways from Brian Chesky
Chesky's hiring framework is direct, no-nonsense, and more relevant to startup founders than anything I've read in an HR playbook.
2 min read
conjoined triangles
thoughts on software, systems, and good practice
What enterprise compliance actually requires from an engineering team
Compliance requirements land on engineering teams whether they asked for them or not. Here's what that actually looks like.
3 min read
From Head of Engineering to CTO in 13 months
What changed. What being CTO actually means at this stage of the company.
3 min read
Cutting operational costs 30% while growing system usage 50%
Specific decisions. Not principles.
4 min read
GCP + Cloud Run + FastAPI: the stack I keep reaching for
Why this combination. What it costs. What breaks. What I'd change.
4 min read
The OKR system that got us to 97.6% completion
I've seen OKRs done badly. Here's the version that worked under real pressure.
3 min read
Multi-agent systems: what actually works in production
Not the research version. The version that has to not fail inside a real product.
3 min read
Growing from 2 to 30+ installations in twelve months
The engineering decisions that scaled. The ones that didn't.
3 min read
YOLO on physical infrastructure
Waste monitoring in the field. Cameras, weather, variable light. The real world is messier than the dataset.
3 min read
What I found when I joined Wasteer's engineering team
Product directing tech. No ownership. Three-hour meetings. The first 30 days.
3 min read
What three years of running CV in production actually teaches you
Not the architecture. The judgment. The things you can only learn by being on call when it breaks.
3 min read
Technical due diligence from both sides of the table
I've been diligenced and I've diligenced others. What the process actually surfaces.
3 min read
2023: the year the ceiling came off
3 min read
AI-assisted sprint planning: the actual workflow
3 min read
Multi-modal models and what they mean for computer vision products
GPT-4V launched. After three years building computer vision products, here's what I think it actually means.
3 min read
Prompt engineering is engineering
The dismissal is wrong. Here's why prompt engineering is a real engineering discipline.
3 min read
Benchmarks vs. production: why they tell you different things
MMLU scores don't tell you what happens when your prompt is ambiguous at 11pm and the model confidently gives you the wrong answer.
3 min read
Why I switched from GPT-4 to Claude for internal tooling
Not a hot take. A practical decision made after running both in production for the same tasks.
3 min read
The compliance problem with LLMs
Our enterprise clients can't send data to OpenAI. For several of them, it's a legal constraint. Here's how we solved it.
3 min read
AI-assisted code review: the system we built and what it actually catches
Six months of AI-assisted code review in production. The architecture, the results, and the things it still misses.
4 min read
LLMs as infrastructure, not features
Bolting an LLM onto a product as a feature and announcing 'now with AI' is not a strategy.
3 min read
Six weeks of GPT-4 in our workflow: what changed and what didn't
Early API access to GPT-4. Six weeks in, here's the honest account.
3 min read
ChatGPT launched. Here's what I actually think.
Having run GPT-3 in production since mid-2020, here's an honest read on what ChatGPT actually is.
3 min read
Speed as a product feature
0.5 seconds average response time in logistics. Not a benchmark — a business requirement.
3 min read
European Patent EP 4 323 939 B1: what we built and why we protected it
What the patent covers, why we filed, and what the process actually looks like from inside a startup.
4 min read
Selling AI to enterprise: what technical founders get wrong
Closed pilots with LVMH, Lego, Moncler, Puma, and Hugo Boss. What we got wrong before we got it right.
4 min read
Building ML infrastructure for under EUR 100K
Enterprise computer vision at production scale does not require enterprise infrastructure budgets. What we built and what it cost.
3 min read
Getting to 99.6% accuracy in production
In a warehouse, under fluorescent lights, with a conveyor belt at operational speed. What it took.
3 min read
Classification vs. detection vs. segmentation: choosing right the first time
The wrong choice costs you six months. How to not make it.
4 min read
Building for millions of requests: the architecture decisions that held
When request volume stopped being a number we celebrated and started being a number we engineered for.
4 min read
Computer vision in the logistics industry: the specific problems nobody writes about
The papers show clean benchmark datasets. The logistics environment is not that.
4 min read
Raising EUR 6M: the technical founder's fundraising notes
What closing a seed round looks like from inside it, as a technical founder who had never raised before.
3 min read
Why we chose YOLO for anti-counterfeiting
The architecture decision that comes up in every investor conversation, written down clearly once.
3 min read
Founding Countercheck: what the first 90 days actually look like
Three months in. What actually happened, before the story gets cleaned up in retrospect.
3 min read
Tim Draper's 'Hero Mindset' - what it means after two years of watching it work
Two years watching the Hero Mindset framework applied by founders. What it means in practice.
3 min read
50 conversations before writing a line of code
How we validated Countercheck before building anything — and what 50 customer conversations actually produces.
3 min read
The EIR model: what it is, how to use it, and when to leave
Six months as an Entrepreneur in Residence. What the model actually is, from someone who's done it.
3 min read
GPT-3 and the moment the ceiling got removed
Early API access. Here's what it actually does.
4 min read
Moving a 500-person global program online in three weeks
COVID hit. Three weeks to move a residential global entrepreneurship program fully online without losing quality.
3 min read
Anti-counterfeiting as a computer vision problem: early notes
Mapping whether computer vision is the right fit for automating product authentication at scale in logistics and retail.
3 min read
The decade in AI: what actually changed between 2010 and 2019
The hype arc of the 2010s, what actually changed, and what the predictions got badly wrong.
4 min read
What Zillion Pitches taught me about building with AI
Two years, several hundred pitches analyzed. What we actually learned about building AI products.
4 min read
Aaron Levie on building Box: the distribution insight nobody talks about
Aaron Levie on why distribution is not a post-product problem — and how Box figured out enterprise go-to-market.
4 min read
Why most AI features in startup products are theater
AI theater: when a product uses AI to perform intelligence rather than to produce it.
4 min read
Tony Hsieh on culture: notes from a Draper session
Culture is what you tolerate, not what you declare. Notes from Tony Hsieh's session with founders.
4 min read
What Phil Libin talked about - and why product longevity is harder than product growth
Phil Libin on asymmetric retention — why some products get more valuable the longer you use them.
4 min read
GPT-2 dropped. Here's what I think it means.
OpenAI didn't release the full model. The stated reason was misuse. Here's what I think it actually means.
3 min read
Speech-to-text in production: latency, accuracy, and the tradeoffs nobody documents
A year of running STT on founder pitch recordings. What the benchmark numbers don't tell you.
4 min read
NLP in 2018: mostly heuristics with a transformer on top
BERT dropped in October. What it changes for practitioners doing NLP in production.
4 min read
What Biz Stone talked about - and why it stuck
Biz Stone on taste as a product skill — and why the discipline of removal is harder than addition.
4 min read
Draper University: what it looks like from the inside
Six weeks, 60 founders from 40 countries, and what actually changes people — from the program side.
4 min read
Sentiment analysis in production: the things the papers don't mention
The paper says 92% accuracy. The production system says something different.
4 min read
Building pitch analysis with AI: what the algorithm can and can't tell you about a founder
Several hundred pitches analyzed. What NLP actually surfaces — and where the ceiling is.
4 min read
IBM Watson: the gap between the demo and the integration
Three months building on Watson for Zillion Pitches. The gap between the demo and the integration is real, and nobody talks about it.
4 min read
San Francisco, 2018: What the City Taught Me About AI That No Conference Could
The ambient conversations alone changed how I thought. You can't get that remotely.
2 min read
Deep Learning Became Boring - and That's the Most Important Thing That Happened This Year
When a technology becomes boring, it becomes useful. Notes on the maturation of the stack.
2 min read
Why I'm Going to San Francisco
The pull of the AI wave. The decision to go be closer to where things are happening.
2 min read
One Year as a First-Time CTO: The Honest Retrospective
What I built, what I broke, and the one decision I'd reverse.
2 min read
Expanding to the US With No Network and Too Much Confidence
Turkey to Silicon Valley. What the ecosystem actually looks like when you're an outsider.
2 min read
The Gap Between "AI-Powered" and AI That Actually Works
2017. Every startup deck says AI. Almost none of them mean it.
2 min read
Azure Cognitive Services in 2017: An Honest Review From a Startup CTO
We used them in production. Here's what the docs don't prepare you for.
2 min read
Shipping to the App Store and Google Play Simultaneously: What Breaks, What Holds
Not a tutorial. A survival guide for doing both platforms at once with no dedicated mobile team.
2 min read
What They Call You When You Do Everything: On Startup Titles and What They Actually Mean
The gap between writing code and leading technical direction is wider than it looks.
2 min read
The Microsoft BizSpark Programme: What $120K in Azure Credits Actually Gets You
Free cloud infrastructure is a gift. It's also a trap if you're not careful.
2 min read
Winning Y2FI and What Nobody Tells You About Startup Competitions
We won and immediately started making expensive decisions.
2 min read
Building a Social App Solo: Xamarin, iOS, Android, and a Backend, All at Once
One engineer. Two platforms. One backend. The decisions that kept me sane.
3 min read
TensorFlow 0.x: First Impressions From Someone Who Had No Idea What a Computation Graph Was
Google open-sourced it two months ago. I've been poking at it since. Here's where I am.
2 min read
The Year I Stopped Being Afraid of Machine Learning
A note from someone who was convinced ML was for PhDs. It isn't.
2 min read
Why I Rewrote My MSc Thesis Project Three Times
Automated testing for mobile apps that generate sensory data. The problem kept reshaping itself.
3 min read
Understanding Backpropagation Without the Math Degree
I spent three weeks on this. Here's the version I wish someone had written for me.
3 min read
My First scikit-learn Classifier and Everything I Got Wrong
Built a text classifier for my MSc thesis. Got 94% accuracy. Shipped garbage.
3 min read