Making Frontier AI Useful
Jan Roose, AI Product Director / Creator
I make frontier AI useful to people by building products, from top-level strategy to production, collaborating closely with researchers and developers
Currently, Head of Product (Fractional) at Dine4Fit (Kaloricke Tabulky).
Most recently, we shipped AI photo logging, helping millions track their energy intake. Now we’re building an AI health assistant.
The work below spans natural-language and image search for the world’s biggest music catalogues, an AI Dungeon Master prototype, and AI-powered satellite imagery analytics well before today’s mainstream models.
Consulting Services
01 / STRATEGY
Building Moats
As software becomes incredibly cheap to build, companies will need to build their moats around data, habits, communities, ecosystems, and brands.
I can help make your product resilient in the age of AI by building moats that retain your current users and help your user base grow.
Together, we’ll review or create your product strategy and roadmap, and I can also support you and your teams with execution.
02 / DELIVERY
Shipping Products
Anyone can now build anything, and the market is flooded with new software. The differentiators are now ambition, judgement, and taste.
With more than a decade of experience across startups and scaleups, I can help you build your next product or feature.
We’ll make sure there’s a solid understanding of the user problem and prepare the right specs for AI agents or developers to build from.
Turning a Calorie Tracker Into an AI Health Assistant
Calsia is the #1 Health & Fitness app in CEE, used by millions of people every month. In under 3 years as Head of Product (fractional), I’ve helped the company multiply its ARR, grow key metrics and add AI capabilities.
#1
Health & Fitness app in CEE
Millions
users
3 yrs
as Head of Product
Building an Intelligent Layer on Top of User Data
We’re building an in-app chat: an intelligent layer that works with their data and gives personalized nutrition advice. Behind the scenes, we designed it with full control over token costs to ensure a positive ROI on the initiative.
Brought AI Meal Recognition to Market in 2024
When the first mass-market image models arrived, we didn’t wait to see what others would do with them. We shipped AI-powered photo recognition for food logging, one of the first products of its kind at this scale in the region.
Over time, we rapidly improved recognition accuracy and grew conversion into subscriptions.
Rebuilt the Experience Without Losing Loyal Users
Calsia has been on the market since 2015 and has built a huge, loyal user base. In 2025 we redesigned the entire app, with new onboarding, new gamification and most user flows optimized, while respecting what existing users already knew and loved.
We significantly grew both early D1 retention and D30 retention.
Built Moats Through Physical-World Integration
The strength of the brand was crucial to keeping the #1 spot on the market. We made deals that brought Calsia into the physical world with features that are hard to copy, like scan-to-log QR codes in restaurants and trains, special B2B deals, and integrations that connect user data to hospital apps and health professionals.
Created the First AI Platform for Space Imagery Analytics
SpaceKnow was running AI-powered satellite imagery analysis long before AI became a major trend. In 2016 I joined to build a platform that made its capabilities accessible to everyone.
1st
AI platform for space imagery
$4M
Series A it helped raise
2016
joined, before the AI wave
Launched a Web-Based AI Platform
SpaceKnow Guardian lets users search and analyze satellite imagery using AI. We acquired early users, and the platform was one of the major factors in raising the $4M Series A.
Defined the UX Standard for Space Imagery Analytics
There was no other product like it. We interviewed people from defense, agriculture, insurance and finance, tested the interface with them, and shipped a UX that still holds up almost a decade later and was copied by competitors.
Launched AI Music Search Used by the Biggest Music Brands
AIMS builds AI models used to search the music catalogs of the biggest music brands, like Universal, Warner Chappell, BMG and APM Music.
Many
major music catalogs
#1
in production-music AI search
2
AI search products shipped
Shipped Natural-Language Music Search
For a long time, AIMS focused on automatic tagging and search by sound similarity. I helped them launch Prompt Search, a completely new way to search for music using only descriptions and natural language. It was sold to many existing and new customers, cementing AIMS as the leader in the production music industry.
Built a Platform for Showcasing AIMS AI Capabilities
After user research with people from the production music industry, we shipped an easy-to-use public platform where anyone can try Similarity Search, Prompt Search or AIMS auto-tagging.
Empowered Sales Teams with AI Visualizations
Instead of explaining our models in slides, we demoed them running inside a version of the customer’s own web app. That way, a lead could see what AI search would do to their own catalog before committing to anything. In an industry with multi-year sales cycles, it moved conversations from abstraction to evaluation much earlier.
Built a Gaming Startup to an $8M Valuation
As a lifelong fan of Dungeons & Dragons, I founded Dungeon Realms, an app that lets people play in a fantasy-themed group chat, not only at a table.
$8M
peak valuation
$1.5M
raised from VCs and angels
3K
Kickstarter backers, $65K raised
Built a Pre-GPT AI Game Master
We fine-tuned an LLM that let people play an adventure with an AI Game Master. This was before the current AI era, and the technology wasn’t ready yet. We briefly worked on it with Tomas Mikolov, the AI scientist behind Word2Vec.
Raised $1.5M at a Peak Valuation of $8M
I raised $1.5M from VCs and angels, including Split82, a co-creator of Beat Saber, the VR hit acquired by Meta.
Crowdfunded $65K from 3K Backers on Kickstarter
We used Kickstarter as a market validation tool and raised $65K from almost 3,000 backers, the 5th biggest Kickstarter success in the category at the time.
Clients & Collaborations
Jan Roose © 2026