About
A small team that ships.
Rezify is a bootstrapped applied-AI studio in Gothenburg. We build consumer products for studying, job hunting and everyday life, and we train the speech and vision models underneath them ourselves — because the off-the-shelf ones weren’t built with our users’ languages in mind.
2025 — a Swedish AB, based in Gothenburg
Two founders — product, engineering and research in house
Bootstrapped — no external capital, lean by design
Build the thing that solves the problem.
Idea to validated product in weeks, with written kill criteria for every bet.
How we work
We don’t over-engineer. We ship something small and real, put it in front of people, and let their behaviour decide whether it deserves more of our time. Anything that doesn’t earn it gets cut on purpose rather than quietly abandoned.
Minnefy and Talanzo both came out of that loop — problems we hit as students, and then again as professionals.
Where AI comes in
We treat models as something to build with, not something to rent. Working from open weights means we can finetune for the languages and markets the mainstream underserves, run inference on our own terms, and keep what we train.
That research isn’t a side project — Swedish speech recognition, low-resource language models and 3D reconstruction all feed straight back into the products.
What we build with.
Full-stack in house — from the app layer down to the training runs.
Apps people actually use
Web and mobile clients, built and shipped by the same two people who decide what goes into them.
- React
- React Native
- TypeScript
- Firebase
- Cloudflare
Speech and language
Whisper-family finetuning for Swedish and low-resource languages, plus the agent and retrieval plumbing that puts those models to work inside the products.
- PyTorch
- Whisper
- KB-Whisper
- LangChain
- LangGraph
Reconstruction and imaging
3D Gaussian splat reconstruction of real environments, alongside applied work on detection and image-generation finetunes. Trained end to end on NVIDIA GPUs — local and rented.
- COLMAP
- gsplat
- CUDA
- YOLOv8
- SDXL + ControlNet (LoRA)
Who’s behind it.
Two founders in Gothenburg. Everything you see was built by them.
Rezify is two problem solvers based in Gothenburg. As students we kept hitting the same practical problems, and as professionals we still do. Most of them aren’t flashy enough for anyone else to chase — but they’re real, and they’re worth solving properly.
Between us we speak Swedish, English, Dari, Farsi, Pashto and Hindi, which is a large part of why the language gap is so obvious to us in the first place. It isn’t a market we researched; it’s one we live in.
Get in touch
Questions go straight to the founders — no forms in between, no support queue.
Email us
hello@rezify.se
Have a look
The products are the argument.
Two are live and free to try. One is launching, one is in build. If you want the longer version, the pitch deck covers the thesis, the research track and where this is heading.
- Live Minnefy — minnefy.se
- Live Talanzo — talanzo.se
- PDF Pitch deck — the longer version