Why I built TapeSave.

The box was in the back of my parents' garage in Louisiana, behind two folding chairs and under a half-empty paint can. Forty-three tapes — VHS, VHS-C, and a few Hi8 cassettes from the camcorder my dad bought in 1989. My mom had labeled most of them in the same blue ballpoint pen across two decades: Christmas 1991. Recital. Trip to Outer Banks '94. Birthday.

He was 76. He had a stroke nine months later. Neither of those things had anything to do with the box. But the order matters, because I'd been meaning to digitize the tapes for two years and hadn't. After the stroke, I drove down on the next free weekend and put the box in my trunk.

Six weeks later a USB stick arrived from Legacybox with 43 MP4 files on it. The total runtime was just over 87 hours.

I copied it. Once. Then it sat for thirteen months.

I'm a cardiologist. I run a clinical practice, I build software at night, and I have three kids under ten. The plan had been to spend a Saturday going through the files, finding the moments that mattered, and sharing them with my mom and my siblings. The plan was always to spend a Saturday on it.

What I learned the third or fourth time I opened the folder is that you cannot, in any meaningful sense, find a moment in an 87-hour video archive labeled Tape_07.mp4. You can scroll past the first ten minutes of any one file, see something you don't recognize, and close it again. You can do that nine times before you give up and watch Netflix.

I tried iMovie. I tried DaVinci Resolve. I imported one tape, dragged the playhead through it, cut at every scene change, named each cut, exported each piece. It took about an hour and I came out the other side with seventeen short clips. The third clip in was 90 seconds of my dad in 1992, taking off his glasses and laughing at something my mom had just said off-camera. I hadn't seen him laugh that hard in a decade.

I texted it to my brother. He replied within thirty seconds.

42 tapes left. 42 hours of manual editing.

I want to be specific about what happened next, because it's what every family digitization project hits and what nobody talks about in the marketing copy of mail-in services. I sat down to edit the second tape the next weekend. I got through it. I sat down for the third the weekend after that. I got through half of it. The fourth tape sat untouched for a month, then two.

The problem wasn't that I didn't care. The problem was that the shape of the work — open a 2-hour file, scrub through it, cut at every scene change, name each piece, set the metadata so it sorts into the right year in Google Photos, export, repeat forty more times — is the shape of work that even very motivated people do not actually finish.

I started looking around for a service that handled the post-digitization step. Mail-in services don't. Legacybox, iMemories, Costco/YesVideo, EverPresent — they all stop at delivering the long MP4. Every one of their websites contains some version of the line "and now you can finally enjoy your family memories!" written above a stock photo of a happy family gathered around a TV. None of them mentions that the file you get back is a four-hour blob that nobody in the family will ever watch.

There was DIY software — PySceneDetect, FFmpeg, a couple of paid video-editing tools with batch features. They worked, sort of, for technical users. None of them was the thing my mom could hand to a friend. None of them dealt with the most important piece of data hidden in every Hi8 and 8mm and MiniDV tape, which is the camcorder's on-screen date burned into the corner of every frame. Those dates are the difference between an archive you can navigate and an archive you can't.

The actual problem was a software gap, not a service gap.

I started building TapeSave on weekends and evening hospital call. The core idea isn't complicated: take the long MP4 a digitization service delivers, run scene-change detection across it, OCR the on-screen camcorder date in each scene, cut the file at the scene boundaries, write the OCR date as file metadata so the clip lands in the right year in any cloud photo app, and hand the family back a folder of individually-dated short clips.

The hard part wasn't any single piece — it was getting all of them right at the same time. Scene detection on home video is harder than scene detection on a movie, because the "scenes" are often just the camcorder being turned on and off rather than clean cuts. Date OCR is unforgiving — a missed digit means a clip from 1992 lands in 1922. The two have to work together, the date has to constrain the cut, the cut has to constrain the date, and the whole thing has to be fast enough that a family can upload a 2-hour tape and get clips back in minutes instead of overnight.

I'm an okay programmer; I've been shipping software since adolescence. I'm a better physician than programmer. The thing that made TapeSave possible at all is that the kind of model architecture you need for this used to require a small team and a million dollars of GPU time, and in 2025 it requires a single person with reasonable engineering taste and a credit card. The cost-curve collapse on consumer ML in the last two years is the quiet thing that lets one cardiologist ship a real product between clinic days.

What I got wrong on the first build.

The first version of TapeSave shipped with what I thought was a clever feature: AI-generated captions for each clip. The model would watch the clip, identify what was happening ("a child opens a gift while adults laugh"), and write a plain-English scene description.

Customers hated it. Specifically: the captions were technically correct and emotionally hollow. "A child opens a gift while adults laugh" is exactly the wrong description of a clip your brain wants to label "Susanna's first Christmas, the year before Grandma died." The captions made the archive feel less like a family memory and more like a stock-footage catalog. The feedback was quick and unanimous — people deleted the captions before sharing.

I cut the feature. The right unit of personal data, it turns out, is not the description — it's the date. The date plus the ten seconds of footage is enough; the family supplies the meaning. Anything else is software pretending to do a thing software can't do.

The number that surprised me.

We track a metric internally that's become the most important number in the business: days from delivery to first family share. The premise is that if a customer gets their clips and within a week has shared at least one of them with a family member, the archive has crossed the line from "digital backup" to "active family memory." That crossing is what the whole project exists to make possible.

For customers who use a mail-in service alone — get back a USB stick of long files — the rate of first-share-within-a-week is under 10%. For customers who go through the splitting step, whether they do it manually or with us, the rate is over 70%. That is the post-digitization gap, quantified. The thing every family thinks they'll do on a Saturday is the thing 90% of families never do.

The TapeSave bet is that turning that 90% into a 30% is worth $9.99 a tape to enough families to support a small, sustainable business. The hospital shifts pay the bills; TapeSave pays for itself; and at some point a few years from now, the box in someone's garage gets opened and the right clip ends up at the right funeral or the right anniversary instead of sitting on a USB stick.

What I owe my dad.

My dad is alive. He recovered most of the way from the stroke; the part of him that laughs at my mom is the part that came back first. Last summer at his birthday I played the 90-second clip from 1992 on the living room TV. He watched it once. He asked me to play it again.

That clip exists in shareable form because of about an hour of work on a Saturday morning two years ago. The other 86 hours of footage from that same box now lives as 1,150 dated clips, sorted into a family album my siblings, my mom, and my dad can browse from any device any time. Most of them are not particularly important. Some of them are the kind of important you can't describe without sounding sentimental.

The honest pitch for what TapeSave does, and what every family should do whether they use us or not, is this: the part of the project that turns a saved archive into a watched archive matters more than the part of the project that everyone already focuses on. The digitization step is necessary. The splitting step is what makes the digitization step worth doing.

If you have a box in your garage, this is your reminder. Find a Saturday. Drive down. Put it in the trunk.

— Phillip Smith, MD
Baton Rouge, Louisiana
May 2026