Where the phone is the pipelineExplore more industriesBeyond the three systems.
Custom AI, built the same way.
Some problems are not a front desk, a pipeline or a back office: product imagery, matching people to schedules, screening a thousand CVs, putting one video into ten languages. We take these on as custom builds, scoped like everything else: one job, one way to check it, one person who owns it.
Product imagery and personalization.
Shoppers expect a realistic preview before they buy, and a photoshoot for every colourway, season and setting is slow and expensive.
- New scenes from a few photos. We fine-tune a generative model on a handful of shots of your product, then generate it in any setting from a prompt.
- A customizer shoppers can talk to. Color, finish and design changes from a plain instruction, rendered on the real product.
- Recommendations from behavior. Products suggested from what a shopper actually looked at and kept.
- Virtual try-on where it helps a shopper decide, so fewer orders come back.
How we check it: every generated image is compared with the real product for shape, color and label before it can be published.
Matching and scheduling. The right person, at a time that works.
Matching a student to a tutor by subject, learning style and availability takes hours by hand, and so does grouping students fairly and filling the timetable around everyone's limits.
- Learns best
- Worked examples
- Free
- Tue, Thu after 4 PM
- Goal
- Exam in March
A. KhanBusy
An LLM-based recommender reads both profiles, so a match explains itself: subject, style and time.
Students grouped by availability and capacity, with session limits respected and conflicts across courses avoided.
Slots filled around availability, workload and priorities, and refilled when someone cancels.
CV screening and ranking. The model reads, a recruiter decides.
A single role can draw hundreds of CVs, and screening them by hand is slow, inconsistent and open to bias. Every CV is parsed, from PDF, Word or a photo of a printed page, ranked against the job description and shortlisted with the reasons shown.
- Parse
- Extract
- Rank
- Shortlist
Meets every must-have; audit depth matches the role.
92ShortlistStrong close process; less audit than asked for.
88ShortlistRight skills, US qualification still pending.
79ReviewBelow the experience floor; flagged, not rejected.
64ReviewNames and photos hidden during ranking
How we check it: rankings are tested against a recruiter's own past shortlists before they are used on a live role.
The same engine, for candidates: a CV improver where two agents critique and rewrite a draft, a CV tailored to a specific job, and job recommendations scored against a profile.
One video. Every language your audience speaks.
Dubbing studios and translators slow a global release down. The pipeline extracts the audio, transcribes it, translates it and voices it again, in sync with the speaker or the character.
- 00:00:01.2 → 00:00:02.8Bienvenidos.Voice matched
- 00:00:03.0 → 00:00:06.4Hoy vamos a ver cómo funciona.Voice matched
- 00:00:06.9 → 00:00:08.6Empecemos.Voice matched
- ExtractAudio from the source video
- TranscribeSpeech to text, with timings
- TranslateInto each target language
- VoiceText to speech, matched to the speaker
- EmbedNew audio and subtitles back into the video
How a custom build runs. The check comes first.
- 1Scope the job
One task, the data it needs and how we will know it worked, written down before anything is built.
- 2Agree the test
A test set and a pass mark from your own examples, so "good" is not a matter of opinion.
- 3Prototype on your data
A working version on a sample of your data, judged against the test before the full build.
- 4Build and hand over
Production build, monitoring and full ownership: the code, the prompts and any fine-tuned model are yours.
Not sure it needs custom AI? Start with a Systems Diagnostic
Before a custom build. What owners ask.
The questions owners ask before they scope a custom AI build with us.
Ask yours on a callYes, when the job is clear and there is a way to check the result. Product imagery, matching and scheduling, CV screening and dubbing are the custom builds we take on most.
If the job is really a front desk, a pipeline or a back office, we will point you to the system that already does it.
You do. The code, the prompts, the workflows and any model we fine-tune on your data are handed over at the end, with the documentation to run them.
We agree a test set and a pass mark from your own examples before we build, and the prototype is judged against it.
If it does not pass, we say so before the full build starts, not after.
It depends on the data and the test, so we do not promise a date until the job is scoped. You get a fixed scope, a fixed price and a date before the build starts.
Describe the job in one sentence.
Write down the task you want a model to do and how you would check it did it well. Bring that to a 30 minute systems call. We will tell you whether it needs custom AI, one of our three systems or no AI at all.
You are on the list.
We will email you times within one business day.
3 leaks found