01

The problem

Every report began as an Excel workbook. Someone rebuilt it by hand into 40–60 slides with up to ~290 charts, then spent a week on revisions: the same fix on twenty slides, five to twelve versions per deck.

02

The product decision

Don't automate slides, automate the source. Templates were derived from 30+ finished decks, so every chart and table is generated from the workbook, and every revision goes to the source and flows to every slide.

03

What we built

Excel in, branded deck out: charts, tables, takeaways, group legends · revisions by plain-language command or by notes on the PDF · diagrams drawn in Excel moved to slides one to one · PDF for the client and editable PPTX.

04

What changed

About 10× faster end to end. The analyst checks the story the numbers tell instead of retyping them, and slides can no longer drift out of sync with the workbook.

01

The problem — two sides

The owner had GPU servers standing idle: a cost with no revenue. Companies wanted LLMs but couldn't send sensitive data to foreign clouds, and didn't want a subscription for occasional tasks.

02

The product decision

Don't rent out servers, sell tokens. A catalogue of open models, each priced per token: clients pick the right model per task, the owner fills the GPUs with many small loads.

03

What we built

Model catalogue and one API · per-token billing, no subscription · isolation that keeps data inside the jurisdiction · load balancing across the owner's servers.

04

What changed

Idle hardware became a revenue line at 80% load. Clients got LLMs for sensitive data without moving it abroad, and pay only for what they use.

01

The problem

Airlines search for pilots through agents, spreadsheets and chats. Pilots fill in the same forms again and again, and good matches are found late.

02

The product decision

Let a vacancy go live from a few inputs that matter — total and night hours, base, rank, type ratings — and let AI do the first screening, so people only talk to likely matches.

03

What we built

Pilot profiles · vacancy builder · AI matching and shortlist · the flow from first contact to interview.

04

What changed

Coming soon: results after go-live.

01

The problem

Two fintech brands needed 63–69 posts a week across LinkedIn, Instagram and X for eight weeks. A studio shoot for that volume is slow and expensive, and stock breaks the brand.

02

The product decision

Build a factory, not a campaign. Start from research into what already works in the category, then give both brands one fictional hero and one visual world, each in its own palette.

03

What we built

A pipeline from brief to channel-ready units · persistent cast, locations and objects · videos up to 20 s with voice and music · carousels and covers in four styles from one text.

04

What changed

From one brief: 13 finished units in a day, each built on a format with a known result. At pilot scale the plan is 504–552 units in eight weeks, with the creative director deciding what ships.

01

The problem

A large outdoor expo of country houses: many exhibitors spread over a big site, and visitors who arrive with a concrete question — who builds, who installs windows, who can help with heating or the garden.

02

The product decision

Put one host at the entrance who knows every exhibitor. A digital person on a vertical screen that you simply talk to: no app, no menus, no queue for a consultant.

03

What we built

A generated avatar and scenes · voice dialogue on an LLM, with the exhibitor catalogue as its knowledge · suggested questions on screen · an interactive venue map that shows the way.

04

What changed

One question gives a visitor an answer and the exhibitor's spot on the map. Exhibitors get the people who are actually looking for them.

01

The problem

Soft skills are learned by doing. Courses give theory, a trainer's hour is expensive, and a colleague to practise on is rarely free.

02

The product decision

Turn every skill into a short scenario with an AI character who reacts like a real person — a demanding client, a nervous candidate, an upset colleague.

03

What we built

A mobile app · AI characters and avatars · scenario-based chat bots for practice · feedback after each conversation.

04

What changed

Practice is there any time and as often as needed. We built it as co-founders, from the first idea to the app in users' hands.

01

The problem

The format was returning to prime time on a fixed date, produced without the original software. Live TV allows no second take.

02

The product decision

Treat it as broadcast equipment, not an app: one system for the operator console, game logic and on-air graphics, with fail-safes tested before every recording.

03

What we built

Operator console · game logic engine · on-air graphics · redundancy and rehearsal mode.

04

What changed

On air since April 2023, four seasons and 50+ episodes without a single failure. Nominated for the national TV award in 2023 and 2026.

01

The problem

Flights happen minutes apart in front of a crowd. Scores from several judges must be combined and shown immediately, outdoors, with no chance to redo.

02

The product decision

A purpose-built scoring tool for the field instead of spreadsheets: each judge enters scores on a device, totals compute instantly, screens and broadcast read the same source.

03

What we built

Judge app · scoring engine · live leaderboard output for screens and the host.

04

What changed

[Result to confirm: teams scored, uptime, whether it was reused.]

01

The problem

A new kind of attraction with no reference to copy, a large public budget and a state expert review to pass.

02

The product decision

Mixed reality instead of pure VR: the physical set, props and effects carry the immersion, so content and hardware are designed as one product.

03

What we built

Realistic content that continues the set · wireless VR for free movement · rain, wind, heat and cold, haptic vests and guns, a tilting platform, players' own hands in the game.

04

What changed

Passed the expert review; tickets sell out two weeks ahead, every week since opening.

01

The problem

A touring exhibition needs an interactive zone that survives packing, shipping and re-installation, and runs every day without an engineer on site.

02

The product decision

Design the zone as a product, not a one-off: the same content, hardware kit and install procedure for every venue.

03

What we built

Interactive floor projection of moving ants · tracking and projection setup · a repeatable install by one team.

04

What changed

The attraction ran 7.5 years on the Las Vegas Strip (June 2016 – January 2024) and was named "Best Place to Meet Your Heroes" by Vegas Seven.

01

The problem

A huge exhibition centre is only reachable for people who can travel to it.

02

The product decision

Make a familiar face the guide: a pop artist's digital double leads the tour, so the experience feels like a show, not a 3D model.

03

What we built

Integration of the artist's digital double as host of the tours through the space and science pavilions.

04

What changed

Opened to users across the country and presented at national technology forums.

01

The problem

In 2015 there was no off-the-shelf way to play one VR film on 100+ headsets at the same second.

02

The product decision

Build the sync and streaming system ourselves and rehearse it like a broadcast.

03

What we built

A system that streamed one film to 100+ mobile VR headsets and kept them in sync, run on site with the agency.

04

What changed

Every guest saw the new Q7 at the same second, together, in VR.

01

The problem

The same interactive experience had to work in 11 host cities at once, on match days, for full matchday crowds.

02

The product decision

One kit, one set of procedures, tech support in every city.

03

What we built

Football skill games, AR fan masks, photo zones; contractor selection and technical management.

04

What changed

All pavilions ran for the whole tournament.

01

The problem

Indoors there is no GPS, and a live broadcast allows no retake.

02

The product decision

Optical tracking across the whole arena volume, with choreography programmed and rehearsed against it.

03

What we built

Tracking volume · drone choreography development and programming · sync with LED screens, lights and wristbands.

04

What changed

The first drone show at a sporting event in the country, live on TV.

01

The problem

The content had to be scientifically correct and still understandable to any visitor.

02

The product decision

Scientists as consultants from day one, three specialised teams working to one plan.

03

What we built

Software, design and object design for an interactive table on the evolution of the universe.

04

What changed

In daily operation since 2018.

01

The problem

A museum had to fit along a staircase inside a working television centre.

02

The product decision

Tell the story through three people rather than a timeline of devices.

03

What we built

Exhibition fabrication, editorial and media content, led across dozens of contractors.

04

What changed

Open since 2021; hosts lectures and events.

Why paid

You see real work before a big commitment. We see the real problem before promising a date.

What it includes

An audit or a working prototype, a plan, and a fixed price for the rest.