Sebastian Loch runs business development at fatsecret, where he has spent thirteen years building out the largest verified nutrition database in the world — country by country, across sixty-two markets and twenty-six languages. This is a check-in on a conversation we started back in 2024, when he was genuinely worried that the AI rush would strip-mine the proprietary data they had spent twenty years verifying. Three years on, the answer is more interesting than either the doom or the hype: the moat held, and it held because of scale, global reach, and two decades of investment in the API. Sebastian walks through how the arrival of LLMs made ten years of their own computer-vision work obsolete overnight and simultaneously handed them the image recognition engine their developers had been asking for since 2016, why they will use AI to improve every process around the data but never to generate the nutrition data itself, and how vibe coding has driven a hundred- fold increase in daily developer signups at a bootstrapped, profitable, never-funded company. We also go two ways: Sebastian turns the questions back on me about API sprawl, AI slop on the API side, and whether any of it can be rated and ranked.
API Evangelist Conversation with Sebastian Loch on Verified Nutrition Data, Building Image Recognition on Top of LLMs, and a 100x Influx of Vibe-Coding Developers
Conversation
Tell me a little bit about yourself again and what do you do.
Thanks for having me. My name is Sebastian, I look after business development here at fatsecret, and I have been here for thirteen years. We operate the largest verified nutrition database globally. We provide access to verified nutrition data on a per-country basis, and we power the majority of the health, fitness, and wellness applications out there. More recently, a lot of the medical, pharmaceutical, and AI-related applications as it pertains to nutrition come to us as well.
You have a wealth of valuable data, and a lot of work goes into it. Where are you at now that AI is consuming everything?
That is probably our biggest talking point, and where we spend most of our time — understanding how this new world of AI affects someone who owns proprietary data at scale, globally, data that is often unique and only we have. Can you make it work, or do you go under because you are being scraped, people ingest your data, take it, and you never see them again? We started back in 2006, when there was no verified nutrition data, no API, no one trying to verify and publish nutrition data at scale — so we built it ourselves. In 2023 and 2024, when the AI rush came to the surface, we were very, very concerned. Would we lose our edge? Would people take our data and run? Would we have a business in three years? Fast forward three years and things are well. We have managed to keep our edge, and some of the AI providers now come and talk to us and tap into our solutions.
How do you use AI internally, and what solutions have you launched since we last spoke?
For the last ten years, our single most requested API feature was image recognition — take a photo of your plate, identify the food and non-food objects, work out what is in the meal. We had a dedicated machine learning team from 2016, and some large corporates tried the same, but nobody could really nail it with traditional computer vision. Then ChatGPT launched, the world went bananas, and it was just there. Everything we and many others had built for ten years was useless overnight. That was somewhat sad and a massive opportunity, because we could then build our own image recognition engine on top of these LLMs and ship it much faster and at scale. We launched it in mid-to-late 2024: it identifies the ingredients and their weights and matches them against the verified nutrition dataset for that user’s country, including restaurant and branded items. We applied the same logic to natural language processing, so a user can just say “for breakfast I had a bowl of cereal, a banana, and a cappuccino.” That has been tremendously helpful in reducing user friction.
Where do you draw the line on what AI is allowed to touch?
All of our processes are now heavily supported by AI, but we can never rely on any output from an LLM for the data itself. We could never ask an LLM for a given piece of nutritional information — that is not something we have ever done and never will. There are hallucinations, there is bias, and there are large inconsistencies outside the US in particular, because so much of the training data for the big LLMs was English-language and Western-centric. Whenever you look outside the US and outside English, the reliability and confidence levels drop significantly. Given we operate globally at scale, we cannot rely on those outputs by any means. So we use AI to enhance our processes and work smarter, but not for our core operation, which is verifying nutrition data and publishing it. Maybe that gets better over time — I am not sure, honestly — but that is where we are right now.
Has your investment in your API over the years set the table for where you are today?
I think so, to some extent, and so has the fact that we tried to build everything on a global scale. A lot of modern conversation is around moats, and having a niche offering is a bit of a threat in itself if you only deal with one tiny component — for us that is nutrition. But we tried to tackle nutrition at a global scale, across all these languages and countries. Together with the investment we made into our API, that is probably the reason we are still here, at least why we still have a B2B business alongside the consumer offering. All these opportunities AI provided, we tried to tackle head-on and build on top of what already existed. If we had not innovated the way we did two years ago, we might not be here today. But the moat itself is the scale and the global reach. Without that, I do not think we would be talking here today.
Is the vibe-coding shift generally positive for you, and how are you addressing it?
For our Platform API, the number of developers signing up every day has gone almost 100x. It is pretty crazy — the traffic, the interest, the inbound is mind-blowing. How many people are now capable and enabled to build their own health, fitness, wellness, and nutrition app is not something we anticipated by any means. So we built a self-serve checkout where people come to the API, use a Stripe subscription, get their own keys, and get started — no email, no contacting anyone, fully self-managed. We started improving those workflows about two years ago when we saw it coming. We are bootstrapped, profitable for twenty years, never raised money, so we knew that if traffic went 10x we could not cope with the influx. We did not anticipate close to 100x. We built automation into our developer support processes and thankfully were able to meet the demand — and the demand has not slowed down. The single biggest spike came when we launched image recognition.
What is the sub-trend inside vibe coding that hardly anyone is talking about?
It is not just people vibe coding an app and putting it on the App Store or Google Play. People can now build and host their own application. There is a project on GitHub called Sparky Fitness where anyone can host their own nutrition app by downloading the repository and running it locally, and there are tens of thousands of people doing it. You do not rely on a mobile app provider, you do not rely on the App Store or Google Play, and you can enhance it — track whatever you want and tap into verified sources such as us. You do not need to be a seasoned backend developer or a native iOS or Android developer. It only lives on your phone, with data you control, nothing shared with marketing providers, no marketing pixels. It is mind-blowing to see, and it is the part of the whole vibe-coding shift that hardly anyone talks about.
Where does the speed-versus-quality trade-off land when you serve regulated businesses?
It probably depends who you are building for. If your audience is someone vibe coding, it is probably fine. But if you work with heavily regulated businesses, with big corporates, possibly with pharmaceutical companies, you need your checks and balances in place, a tight security framework, and all of that takes time. It slows you down and it does not come overnight. So we are possibly still moving very slowly compared to many other agentic solutions out there — but we are trying to keep a robust offering. We cannot just vibe-code something, put it out there, and have it cause more harm than benefit. It is really a trade-off between speed and quality. Usually you work in a triangle between speed, quality, and cost, but in this instance I would say it is just speed and quality. If you have nothing to lose, hit the throttle and go full steam ahead. Unfortunately that is not us. And I think that is part of why we are seeing this massive influx of APIs out there.
Sebastian Loch
Sebastian Loch heads business development at fatsecret, where he has worked for thirteen years and looks after the Platform API business. fatsecret has been verifying and publishing nutrition data since 2006 — before there was any such thing as a nutrition API — and today operates the largest verified nutrition database globally, covering generic foods, branded foods, and restaurant menu items across sixty-two countries and twenty-six languages. It powers the majority of health, fitness, and wellness applications, and increasingly medical, pharmaceutical, and AI-driven products. The company is bootstrapped and has been profitable for twenty years without venture funding.
