A multi-camera video remix platform
Best viewed on a desktop or tablet. This case study was designed as a presentation; the scroll version is below.
Stitches
Challenge
I had a theory: when people who filmed the same moment get back one remix they made together, they’ll want to share it and do it again.
The usual way to test that is raise money, hire a team, and wait months to find out. I didn’t want to wait, and I didn’t want to spend someone else’s money proving a hunch.
Solution
So I built it myself. Using AI-assisted coding in Claude Code, I stood up the whole serverless pipeline solo, video intake over iMessage through audio sync, scoring, render, and delivery back to the thread. No round, no team, no runway.
Just the fastest path from theory to a working thing I could put in real hands and measure.
Scope of Work
Case Studies
Client
Self-initiated
Year
2026
Case study // Stitches
A multi-camera video remix platform
Self-initiated
2026
Challenge
I had a theory: when people who filmed the same moment get back one remix they made together, they’ll want to share it and do it again.
The usual way to test that is raise money, hire a team, and wait months to find out. I didn’t want to wait, and I didn’t want to spend someone else’s money proving a hunch.
Solution
So I built it myself. Using AI-assisted coding in Claude Code, I stood up the whole serverless pipeline solo, video intake over iMessage through audio sync, scoring, render, and delivery back to the thread. No round, no team, no runway.
Just the fastest path from theory to a working thing I could put in real hands and measure.
The problem
Every shared moment gets filmed by a dozen phones. The school play, the birthday, the trick landing at the skate park.
Everyone captures it from their own seat, their own angle, their own version of being there. Then it ends and every clip disappears into a separate camera roll.
The collective experience happened.
The collective artifact never does. The footage exists, scattered across twenty devices, and no one ever makes it into something whole.
The insight
The output isn’t anyone’s video. It’s everyone’s.
Most tools treat the person holding the camera as the author, so you get Sarah’s video of graduation. Stitches treats the moment as the subject and the people who showed up as collective authors, so you get what ten people made at graduation together.
That single reframe drove every decision that followed. If the moment belongs to the group, the product can’t reward individual performance, can’t run on followers, can’t center one creator. It has to be built so the artifact only exists when more than one person contributes.
The decisions
Three calls carried the build.
First, texting instead of an app: no download, no account, no new behavior. People already text videos, so I pointed that habit at something new and watched if they came back.
Second, a human in the loop before automation: I cut the first remixes by hand to learn what a good one feels like, then built the engine that makes them. Audio sync, action scoring, cuts between cameras, render. The hand-cut path still exists for curated events. The default is the machine.
Third, the platform as a measurement instrument first: every clip, contributor, and reply is captured, because the real question isn’t whether the pipeline runs. It’s whether people who get the remix back want to do it again.
The role
I built this solo. Sole designer, architect, and builder.
I’m a product person, not an engineer by training, and to prove the concept was real I stood up the entire serverless pipeline myself, from video intake over iMessage through audio sync, ML scoring, render, and delivery back to the thread. AI-assisted coding in Claude Code is how I got fluent in territory I’d normally hand to a team.
The work that stayed mine throughout was the work that matters most here: product judgment, system design, and deciding what the thing should feel like.
The tools let me ship it. The decisions were the job.
Every state, one line
The states
Every message the system sends is a state. Every state gets one line.
First contact, once per phone, ever.
Lobby wait. Pairing. Timeout, which doubles as the retry.
Daily cap.
Help, when someone texts words: the only reply a stranger gets, so it carries the pitch. Once a day.
The settings page
All of it in a settings page, not code.
Change a line, the next message uses it.
Outcome
The question
Stitches hasn’t launched. The demo phase exists to answer one question before it does: do people who get the remix back want to do it again?
The demos
The goal was never scale. It was evidence. Live demos in chosen contexts: parents at the same school event, friends at the same bar, a crowd at the same skate park.
The measure
Invited versus uninvited contributors, contributors per event, turnaround time, and what people do when the remix lands. Do they share it. Do they ask to do it again.
Try it
Next time two of you film the same thing, text it here →
The problem
Every shared moment gets filmed by a dozen phones. The school play, the birthday, the trick landing at the skate park.
Everyone captures it from their own seat, their own angle, their own version of being there. Then it ends and every clip disappears into a separate camera roll.
The collective experience happened.
The collective artifact never does. The footage exists, scattered across twenty devices, and no one ever makes it into something whole.
The insight
The output isn’t anyone’s video. It’s everyone’s.
Most tools treat the person holding the camera as the author, so you get Sarah’s video of graduation. Stitches treats the moment as the subject and the people who showed up as collective authors, so you get what ten people made at graduation together.
That single reframe drove every decision that followed. If the moment belongs to the group, the product can’t reward individual performance, can’t run on followers, can’t center one creator. It has to be built so the artifact only exists when more than one person contributes.
The decisions
Three calls carried the build.
First, texting instead of an app: no download, no account, no new behavior. People already text videos, so I pointed that habit at something new and watched if they came back.
Second, a human in the loop before automation: I cut the first remixes by hand to learn what a good one feels like, then built the engine that makes them. Audio sync, action scoring, cuts between cameras, render. The hand-cut path still exists for curated events. The default is the machine.
Third, the platform as a measurement instrument first: every clip, contributor, and reply is captured, because the real question isn’t whether the pipeline runs. It’s whether people who get the remix back want to do it again.
The role
I built this solo. Sole designer, architect, and builder.
I’m a product person, not an engineer by training, and to prove the concept was real I stood up the entire serverless pipeline myself, from video intake over iMessage through audio sync, ML scoring, render, and delivery back to the thread. AI-assisted coding in Claude Code is how I got fluent in territory I’d normally hand to a team.
The work that stayed mine throughout was the work that matters most here: product judgment, system design, and deciding what the thing should feel like.
The tools let me ship it. The decisions were the job.
Every state, one line
The states
Every message the system sends is a state. Every state gets one line.
First contact, once per phone, ever.
Lobby wait. Pairing. Timeout, which doubles as the retry.
Daily cap.
Help, when someone texts words: the only reply a stranger gets, so it carries the pitch. Once a day.
The settings page
All of it in a settings page, not code.
Change a line, the next message uses it.
Key insights
The question
Stitches hasn’t launched. The demo phase exists to answer one question before it does: do people who get the remix back want to do it again?
The demos
The goal was never scale. It was evidence. Live demos in chosen contexts: parents at the same school event, friends at the same bar, a crowd at the same skate park.
Try it
Next time two of you film the same thing, text it here →
The measure
Invited versus uninvited contributors, contributors per event, turnaround time, and what people do when the remix lands. Do they share it. Do they ask to do it again.