What Happens When the Person Who Runs the Court Reporting Tech Also Builds the Tools

My view of legal tech comes from the operating side: sitting in depositions, opening proprietary court-audio packages late at night, and explaining software changes to reporters who still have to finish the record. That experience changes what I build.

By day I’m the Director of Technology at Executive Reporting Service — a court reporting firm that’s been producing the legal record in Florida since 1981. By night I build software for the same industry as DepoStack. Running the technology and building the tools are usually two different jobs, held by two different people who rarely talk. Doing both means I can’t hide from either one. The thing I ship at night, I have to live with the next morning.

The gap nobody admits

In my work, the gaps are concrete: vendor-specific containers and multi-file recording packages that ordinary players may not open as a complete session, delivery workflows built around attachments, and AI features that treat transcription as the whole job.

A transcript is evidence, not generic content. A person remains accountable for every page and line. My rule is simple: use AI to support the reporter’s judgment, not to impersonate it.

The end-to-end digital court-reporting workflow makes those human handoffs and accountability points explicit.

Why I started building instead of buying

That conviction is why I build instead of just buying. Off-the-shelf court reporting technology kept failing the same way: it solved a vendor’s problem, not the reporter’s. So I started shipping the tools I wished existed.

The first was DepoAudio. The pitch is unglamorous and exactly right: court recordings shouldn’t be held hostage by the software that made them. It converts supported SGMCA, BWF, and standard media into standard audio you can actually use, entirely on your machine. FTR/TRM handling is experimental because delivered variants differ, so each result still has to be verified. No subscription, no cloud upload, no account. For a field that handles sensitive testimony, “on-device” isn’t a feature — it’s the requirement. It’s free and open source, because some plumbing should just be plumbing.

The second is CasePrompts, and it’s where the question of AI for court reporters gets interesting. The litigation world is drowning in generic AI prompts that hallucinate confidently and cite nothing. CasePrompts takes the opposite approach: its CITE templates require page:line citations and instruct the model to identify claims the supplied record does not support. Those are prompt requirements, not a guarantee about a model’s output; the result still needs professional review. It’s prompting like you’re assigning a task to a junior associate, not Googling. That distinction is the entire game in legal AI.

Building from inside changes what you make

When you also run the technology, you can’t ship a demo. You feel the format that won’t open, the reporter who needs the tool to be obvious, the attorney who needs the cite to be exact. Court reporting technology built by someone with no stake in the record tends to optimize for the screenshot. Built by someone who has to answer for it Monday morning, it optimizes for trust. Those are very different products, and the field can tell the difference even when it can’t name it.

None of this is anti-AI. I build on these models every day. It’s anti-pretending — pretending the model is the professional, pretending proprietary lock-in is a moat instead of a liability, pretending court reporting is a transcription problem instead of an accountability one. The firms and reporters who come out ahead won’t be the ones who removed the human fastest. They’ll be the ones who got faster without giving up the thing that made them trustworthy in the first place.

That’s the bet I’m making, and I’m making it from inside the building.

That is the bet behind DepoStack, which is currently in early access. Open the current work.