The founders
Three of us. One idea at a time, done well.
Jens G.
Kateryna F.
Mickaël A.
What if the tools an AI studio needed to make films could also run a small business? We spent three years finding out. Turns out: yes, and rather well. This is the story of how a film collective accidentally became a software company — and why that's exactly the right background for building agents that ship real work for real humans.
We met on a film. The usual way: a friend-of-a-friend Discord, a half-written script, an absurd deadline. Jens had the camera eye. Kateryna had the production brain. Mickaël had the ear and a background scoring for European hotels — the kind of places where a voice on the line is a promise of something warmer. Nobody was paying us. We built what we needed.
The first pipeline was stitched together in a week. Whisper for transcription, SDXL for stills, Bark for voice, ffmpeg for everything else. It worked. Night Operator played at East Village NY and we spent the prize money on coffee and a server. By 2024 we had built enough tooling that we started calling it Codec. By 2025 the tooling had become a product — and the product had become a company.
"We didn't set out to build a business. We set out to make films that didn't exist yet. The agents came from solving the same problem twice."
Here's the insight that turned AVA from a film collective into a software studio: the hard parts of making an AI film — keeping a character consistent across 43 shots, routing an event through six tools without losing state, getting a voice to actually sound like someone — are the exact same hard parts of running a small business on top of AI. Continuity. Memory. Voice. Trust.
So when a dentist in Paris asked us whether her front desk could be a voice agent, we already had the answer. When a law firm in Massachusetts asked whether an AI could triage retainers in the partners' tone, we had written that engine for three different directors already. The instruments were built. They just needed to be pointed at a different score.
Jens G.
Kateryna F.
Mickaël A.
Every AVA product started as a fix for something on a film. The compositor became a pipeline. The pipeline became Codec. Build the thing you need; keep it; share it.
When an agent answers in your company's tone, it's not a UI — it's a contract. We spend unreasonable amounts of time on how things sound because every call is the whole brand for someone.
The big firms get Palantir. The corner clinic gets a phone tree. We're fixing that asymmetry, one retainer-drafting agent at a time.
Codec is MIT because it has to be good enough that somebody else wants to ship it. Anything less is marketing. We'd rather lose customers than hide the engine.
Three strangers in a Discord finish a 12-minute film. It plays at East Village NY. Nobody has a company yet. Prize money buys a server.
The pipeline gets a folder, a license (MIT), and a versioned repo. Jens pushes the first commit at 2:47 AM Copenhagen time.
The temporal-state engine behind a 7-minute single take becomes the scene memory module Codec ships today.
A boutique dental practice in Paris asks for a phone that answers in French, English and Italian. We build Intake v0 in six weeks. She still pays us.
Wyoming. Three founders, one client, fourteen months of receipts, a working product. Kateryna registers the .ai domain from a coffee shop.
Lucy, Intake, Beacon, Codec. We have a pricing page. We have a waitlist. We have not yet raised a round.
First institutional round. Goal: hire four engineers, open a US branch, make the waitlist disappear. We'd love to meet.
That's the end of the story part. If you want to put us to work — or just see if we're the right kind of people for what you're doing — we're a call away.
Book a consultation