Sam Baud · Amsterdam · AI for documentary and media teams

Let AI log the rushes, so you can make the film.

Logging rushes, finding the moment in fifty hours of interviews, the paperwork around a film: the tasks that eat a small team's week. I start with one small project, measure it against how you work today, and grow only what works.

Book a free half-hour callA first pilot: €3,500 excl. VAT, fixed.

Data scientist at NU.nl (DPG Media), then co-founder and CTO of Typetone. Admin AI only: no generated footage or voices.

↓ Scroll

Every week someone on your team spends a day logging footage, searching transcripts or filling in clearance lists. That day is where I start.

How I work

Measure first. Start small. Grow what works.

Day oneTalkYour work, and what good looks like.
Week oneThe test firstBefore any AI: your real cases, and what counts as good.
Week twoThe setupThe right tools and context, so the model does its best work.
Weeks three to sixA small pilotOne task, in your real work, scored against today.
If it holdsExpand, slowlyGrow what passes. Keep scoring what is live.

↺ every change is scored again

A first project

One task, a few weeks, a clear answer.

One taskSomething your team does every week: logging footage, searching interviews, a first pass at a paper edit, archive and clearance lists.

Two to six weeksI work one or two days a week, so a project stays small on purpose and you stay close to it.

A fixed priceThe pilot is €3,500 excl. VAT, fixed: about five days of my time over the weeks. The first half-hour call is free.

A clear answerA tool your team uses, scored against today. If it does not beat today, you have spent one fixed pilot, not a year of licences, and you know where AI does not fit.

Where it landsIn the tools you already use: markers or an EDL for Premiere, Avid or Resolve, and a Word or Google doc for a paper edit. The tool and its tests stay in your own accounts.

Your material, your callIf you want, models that keep nothing (zero data retention) or run only in the EU. Either can narrow which models we use, or how good they are.

After the pilot

After the training, who builds it? Trainers leave a team with prompts; I leave a process that runs.

It keeps workingA day a month: watching it, fixing it, small changes, and new models when they help.

A team dayHalf a day for the whole team on their new way of working, with a shared prompt library and an AI-use policy that follows the NVJ guideline.

The next processA second workflow on the same setup, so it costs less than the first.

Per productionThe logging and subtitle pipeline set up for each new film, with check-ins during the edit. It fits in a production budget.

A part-time AI leadTwo days a month once more than one process runs: what to build next, which tools, and the team’s questions.

Admin AI, not creative AI: logging, subtitles, research, applications, a first paper edit for the editor to rework. The editor makes the film. No generated footage or voices.

How we know it works

One question: is it better than today? Before a pilot starts we agree on the number, and we measure it on your own material.

When AI assists · an example

4 h → 1 h 30

Logging one interview to selects: four hours for an assistant editor today, an hour and a half when a model drafts and the editor checks and finishes. We count the time to a result they would sign off.

When AI decides · an example

79 of 100

Clips tagged by location without a person, 5 of them wrong; the 21 it was unsure of went to a person. Every call is checked against answers your team gave.

Talk to mehello@sambaud.com

A free half-hour call, in Amsterdam or online: your work, one task worth trying, and whether a small pilot makes sense.

Try it

Cut drafts a first paper edit of a short documentary; you ask for changes and decide what stays. Watch the edit change, and read why.

CutA first paper edit from four interviews, changed by asking“End on something that lands”

Work, without the names

Inside other companies' products, theirs to announce.

A national newsroomData scientist at NU.nl (DPG Media), working next to the editors.

Marketing content, written by AIAt Typetone, posts and articles drafted in each client's own voice.

Catalogue from the open webProduct pages turned into a clean catalogue, with a flag on anything the model was unsure of.

Where this comes from

Data work at Dynasource and RAI Amsterdam, then data scientist at NU.nl (DPG Media), so I know a newsroom from the inside. Then four years building AI products, most as co-founder and CTO of Typetone. We built Sarah, an AI employee who ran marketing on her own. She worked, but the customers who got real value kept a person in the loop.

That is what I build now: a person and a model, working in turns.

Cost. A tenth of a cent per call decided whether a product could exist.
Test first. A test set before a model gets a task, kept running once it is live.