Home Our work
The problem, and what we built.
Four products and three studios we run ourselves, on the same platform we use for client builds. The films are illustrative.
Product
Conversation
A spoken job. Text to voice and speech to text. It can open a record and hand the moment to a person.
The problemThe job is spoken, but the only tool on offer is a chat box beside it.
What we builtConversation holds a live turn: it listens, it answers from the job you wrote, and a person can take over.
The problemText to voice and speech to text live in different places, so the words and the audio drift apart.
What we builtBoth sit in this product. A person can correct a name, a number or a term before it is stored or acted on.
Product
Documents
Digitise the whole page, or extract the fields you name. Every block can be checked against the source.
The problemScans, forms and handwritten pages become a transcript that scrambles the page.
What we builtDigitise the whole page and keep the reading order, or extract only the fields you name.
The problemA table or a column is dropped into a paragraph and called done.
What we builtLayout stays. Every block can be checked against the place it came from on the source.
Product
Content Studio
Articles, campaigns, scripts and social, then a translation or a narration. Publish waits for someone named.
The problemEvery article, campaign, script or social post starts from a blank page.
What we builtA brief becomes a draft. The team can still edit it. Publish waits for someone named.
The problemThe same piece has to exist in another language, or as something a person can hear.
What we builtTranslation and narration stay with the draft. A reviewer checks the language. Someone listens before it is published.
Product
Model Training
One task on your data. Evaluation first. You keep the weights and can run them on your machines.
The problemThe task has to stay on your data. A shared model is the wrong place for it.
What we builtWe train for one task. You keep the weights and can run them on your machines.
The problemA demo looked convincing, so the model is treated as ready.
What we builtEvaluation is agreed before training. The model does not ship until that evaluation is met.
Studio
Education Apps
An AI tutor, and the videos and content our team makes for the education product. A teacher approves.
The problemA chatbot sits beside the course and is not the lesson the student is on.
What we builtThe tutor explains the stuck step from that lesson. Practice and checks follow the skill, not a random quiz.
The problemThe study path ignores what the student already missed.
What we builtA study-path model chooses what to revise from the misses. A teacher can still change it.
Studio
Graphics Studio
Campaign art, product visuals and brand systems, as files a designer can still edit.
The problemCampaign art, product visuals and brand systems start from a blank canvas every time.
What we builtA graphics app bound to the brand. The next brief starts from a system, not a fresh prompt.
The problemA generated image cannot be opened by the designer who has to ship it.
What we builtOutputs stay as files a designer can still edit in the tools they already use.
Studio
AI Film Studio
Ad films and movies, made by our team — and the videos and content inside an education product. A person reviews the cut.
The problemAn education product needs a lesson on screen, not only a page of text.
What we builtOur team makes the video and the content. A teacher clears the cut before a student sees it.
The problemAn ad film or a movie is needed, and a generated clip is treated as the release.
What we builtThe team writes it, makes the picture and cuts it. A person reviews the cut before it goes out.
The stack behind each one.
Conversation
Open weights, Nvidia, and paid models. We pick the one that holds the voice, the transcript and the cost of this job.
Documents
Language and vision models read the page. The review against the source is the product, not an extra step.
Content Studio
Open weights and paid models for the draft. The person who approves publish is not a model.
Model Training
PyTorch and Nvidia for the run. Open weights as the start. The result is yours, not a model we keep.
Education Apps
The same open and paid models as the other products, inside an app. A teacher approves the path and the cut.
Graphics Studio
Image models on PyTorch and Nvidia, plus paid models when the brief needs them. Designers keep the file.
AI Film Studio
Scripts can start on the same models as Content Studio. The cut is the team’s, and a person watches it before release.
Let’s talk about what you want to ship.
Start with a two-week, fixed-fee discovery. You get a ranked portfolio, an architecture, a cost model and a fixed-price build plan.