How I used AI, a phone camera, and my family to build immersive law enforcement training on a shoestring budget.
Last Tuesday, I filmed my fiancé taking out the trash. Sitting on our porch swing. Walking through the house.
Not for a scrapbook. Not for a TikTok trend.
For law enforcement training on child exploitation cases.
I even filmed my 10-year-old golden doodle, Rey, doing absolutely nothing suspicious.
I know. It sounds strange. But here’s the reality:
1 in 5 children are sexually abused before they turn 18. And in the majority of the cases we see, technology is involved in some way.
So when we build training for investigators and prosecutors, it has to feel real. The footage has to look real. The digital trail has to look real. Because the cases are real.
What’s actually wrong isn’t that we’re filming “fake surveillance.” It’s that the work is necessary in the first place. And until those statistics change, we build better tools. Better simulations. Better training. That's my motto, anyway.
Because pretending this isn’t happening isn’t an option. And honestly? It's stuff like this I'm most proud of from the last few months of work.
Welcome to nonprofit tech. Population: us, and our very confused (but willing) families.
How I ended up as an accidental film director
Here's the thing about building high-stakes training when you don't have a Hollywood budget: you improvise. A lot.
For the past several months, I've been deep inside one of Zero Abuse Project's most intensive programs. Tech Crime to Trial isn't a slide deck. It's not a webinar where you half-listen while checking email. It's a 4.5-day immersive experience where ICAC investigators and prosecutors travel to one desitnation and actually work a child exploitation case from the first tip all the way to the courtroom.
Real teams. Real collaboration. Fictional case, but everything else is as close to the actual experience as we can make it. Participants draft warrants. Run OSINT searches. Execute mock search operations. Interview suspects. Prep for trial.
If they miss something early, it cascades. Because that's what happens in real cases.
My job is to make the fake stuff feel real enough that the training sticks. Real enough that when an investigator is back in their jurisdiction at 2 a.m. staring at a new case, they remember what worked and what didn't.
So when I needed surveillance footage and the budget said "absolutely not" to professional actors, I grabbed my phone and recruited my unsuspecting family. My partner never asked questions. He just took out the trash while I parked across the street & crouched behind the front seat of my mini van like a woman who has fully lost it.
This is what building things that matter actually looks like. Sometimes.
The work nobody talks about: building a fake person's entire digital life
You know how TV writers create backstories for characters? That's part of my job now, except the character is a fictional perpetrator and the "audience" is a room full of investigators trying to build a prosecutable case.
I've spent hours building OSINT profiles. Social media accounts that feel lived-in, not manufactured. Posts that reveal behavioral patterns without being too obvious. Photos, metadata, timestamps. The digital breadcrumbs that investigators actually follow in real cases.
This is also where AI has quietly changed my work in ways I didn't fully anticipate.
I want to be careful here, because this matters: I'm not using AI to teach investigators how to do their jobs. I'm not using it to make case decisions or flag behavior.** I'm using it to scale realism without burning out my small team.**
Building authentic training artifacts used to take weeks. Fabricating mock evidence that felt genuinely realistic meant either outsourcing to expensive vendors (not happening) or having my team manually create everything from scratch (also not sustainable). Now I can prototype faster using AI tools like Genspark. Generate variations. Test different scenario branches without rebuilding everything from zero every single time.
AI gives me scaffolding. Then I focus on the details that actually matter: accuracy, trauma-informed design, and making sure the training reflects what investigators will actually encounter in the field.
It's not magic. It's leverage. There's a real difference.
The thing I'm weirdly most proud of
Honestly, the surveillance footage and the OSINT profiles aren't even the biggest lift. The project that's occupied the most brain space is the Tech Crime to Trial Logistics Portal, a custom web app built in Base44 that acts as mission control for the entire course.
Participant onboarding. Team dashboards. Secure file sharing. Mock evidence storage. Schedules, task lists, chat functions. Admin tools to track who's prepared, who needs what, and how information flows between teams and instructors and mock judges.
Before this existed, the course ran on email threads, printed binders, and sheer determination. Instructors juggling excel spreadsheets. Participants hunting for the right document at exactly the wrong moment.
Now it's one system. One source of truth. It's modern, interactive, collaborative. And it mirrors how modern investigations actually work: digital-first, collaborative, with controlled information sharing and real-time updates.
This is the infrastructure work that nobody sees. When it works, it's invisible. When it breaks, everything falls apart. I rebuilt a broken feature the morning of a training once. I've troubleshot failed uploads at midnight laying sideways in bed. I've done things with duct tape (metaphorical) that would make a developer cry.
That's fine. That's the job.
What I actually want creatives to hear
If you're a creative working in mission-driven spaces, or you're thinking about using AI in your work, here's what I need you to know.
**AI is not the hero of this story. The course is the hero. **
The instructors who built the curriculum over years of real experience. The investigators who show up and do hard, necessary work. The prosecutors who have to carry what they see in courtrooms.
But thoughtful technology, including AI, made it possible for me to do this work without sacrificing quality or safety. It helped me build better scaffolding so these highly-trained humans could do what only humans can do: apply judgment, collaborate under pressure, and make decisions that protect children.
That's the kind of AI use that excites me. Not the flashy stuff. Not the hype cycle. The boring, infrastructure-level work that makes high-stakes training accessible and sustainable on a nonprofit budget.
And the part that matters most to me, the part that I think gets lost in a lot of AI conversations: we were intentional about what AI touched and what it didn't. In spaces where the stakes are this high, where the people using your tools are trying to protect the most vulnerable, "move fast and break things" is genuinely not an option. What breaks is trust. Safety. Justice.
That's a line I won't cross, even when the timeline is brutal and the budget is tighter than it should be.
Coming up for air
The next session is prepped for our March training in Phoenix. The portal is live. The fake evidence is fake enough.
My partner still doesn't fully understand why I asked him to sit outside our house on a laptop and "look creepy." My kids thought it was hilarious. The dog remains unbothered.
And I'm sitting with a lot of questions about what it means to build ethically in spaces this consequential. About how we scale without losing what makes the training actually effective. About the responsibility that comes with building tools at this intersection of technology and justice.
I don't have clean answers yet. But I'm thinking about them, and I'll keep bringing you along for that process.
If this resonated, I'd genuinely love to hear what you're building and what tradeoffs you're navigating. This community is one of the places I think most clearly, and I'm grateful you're in it.
More soon.
Stephanie
_2025.png){kind=logo}