What we build, and what we don't.

1891 isn't a sign-language AI lab. We're an interactive experience company. Here's the line we draw — and why it lets us ship today rather than wait for a frontier to clear.

Every few weeks somebody asks us a version of the same question: "Are you building a translator? A real-time interpreter? A wearable that captions the world for Deaf people?" The honest answer is no, and the longer answer is the one worth writing down once so we can point at it instead of typing it again.

Full sign-language translation — continuous signed sentences with facial-expression nuance and two-handed compounds and discourse-level prosody — is one of the most important frontiers in machine learning. It's also the work of interpreting agencies, research labs, and AI companies with very specific funding shapes. We're cheering for them. We are not them.

The line we drew, and what's on each side

We sat down early and divided the space into three lanes. Here they are:

• What we use Static handshape recognition (A–Z). Number recognition (0–9). Universal gestures like ILY, 👍, 👎. Single-hand motion for J, Z, and simple swipes. Context-aware limited-vocabulary UIs. Fingerspelling-based selection.
• Watching closely Two-handed compound signs at speed. Continuous chained fingerspelling. Common-vocabulary single-sign recognition (~500 signs). Facial-expression markers as grammar. Regional dialect adaptation.
• Not our lane Full conversational sign-language translation. Real-time bidirectional interpreting. Wearable-camera language access. Discourse-level prosody. Cross-language sign-to-sign translation.

The first column is what's on our roadmap today. The second is what we'll bring in when it stabilizes. The third is somebody else's work — and that's a feature, not a bug.

Why this lets us ship

The pieces we actually need — static handshapes, digits, a small gesture set — are already ready. Off-the-shelf. That's enough for thousands of useful interactive moments, and most of the spaces in our lives are missing exactly those moments today.

An ATM that responds to a four-digit PIN signed in the air. A drive-thru menu that watches the hands. A children's-museum guestbook that prints a name signed by a six-year-old in giant typography. A keypad on the back door of the staff lounge that doesn't require a sticky touchpad. A signless, touchless way through ten thousand small interactions per day in a hospital, a school, a hotel, a stadium.

None of those need a continuous translator. They need a camera, a screen, a small confident vocabulary, and the willingness to design around the body that's actually in the room.

Sign language is the user interface no one chose, and the one that's been hiding in plain sight.

The other reason we draw the line here

Translation-grade systems carry consequences a little different from what we ship. If a translator drops a clause, somebody misunderstands an emergency. If our drive-thru menu drops a "B" because the kid signed it too fast, the kid signs it again and laughs. The stakes are different — and the engineering ethics are different.

We're builders. We solder our own boards in Frederick, write our own firmware, design our own gameplay, and ship installations that need to run for years without re-flashing. The work we do should match the part of the field where it can hold up.

Real interpreters are people. The places we want to put cameras don't need an interpreter — they need a moment of welcome. Those are two different jobs.

One more thing

You may have noticed we don't lead with the Deaf-built story. That's deliberate. The brand line — Designed by Deaf people. Built for everyone. — is real, and the family this company is named after has been Deaf since 1891. But the work has to stand on its own merits in front of every kind of guest in every kind of room.

The bullet that does the actual heavy lifting is the one in the third lane: universal design isn't a retrofit — it's a starting point. We design for the body that shows up, not the body we wished had shown up.