Doolittle Corporation’s Rosie Ai Translates What A Dog Is Feeling

(PRESS RELEASE) ST. PETERSBURG, FL — Doolittle Corporation opened public beta access to ROSIE — the Real-Time Operating Service Interspecies Element — an AI engine that reads a dog’s emotional state from the full range of signals the animal is already giving off, and renders it in plain language, in the dog’s own first-person voice.

ROSIE runs free at rosie.doolittlecorp.com. It opens in a browser. There is no signup, no download and no hardware to buy, and anyone can test it in about a minute. The company, incorporated in Delaware in March 2026, is opening the beta alongside a patent-pending filing covering its sensing and emotional-inference architecture, and A Science of the Canine Mind — a two-book work by founder Danielle A. Franklin scheduled for Summer 2026.

Most pet technology reports what a dog did. ROSIE is built to interpret what a dog means.

It started with two languages that had never been written down

In 2022, Danielle Franklin was learning Hokkien Taiwanese — a language with millions of speakers and no standardized written form — while raising her third yellow Labrador. The dog barked, guarded, and pushed back in ways that every training manual filed under disobedience.

Working on both problems in the same season, Franklin recognized she was looking at one problem twice.

“Rosie was not misbehaving. She was speaking a language I did not yet understand. We have spent a century teaching dogs to understand us, and almost no time building the other half of the conversation. A dog who is frightened and a dog who is dangerous can look identical from across a room. The dog knows the difference. We are the ones who cannot hear it.” – Danielle A. Franklin, Founder & CEO, Doolittle Corporation

The engine is named for her.

How ROSIE works — open foundations, a proprietary emotional layer

ROSIE does not rely on biometrics alone. It fuses four channels — body, voice, expression and environment — and weights each by how confident it is in the signal, so a noisy room or an obscured camera degrades the reading gracefully instead of producing a confident wrong answer.

The foundations are deliberately open and citable:

• Audio is classified on-device using Google’s AudioSet ontology via YAMNet, with an ESC-50-trained classifier that separates the dog from everything else in the room.
• BEN, the reasoning layer, retrieves against vetted canine behavioural literature rather than improvising from general web text.
• CER, the Canine Emotional Roadmap, maps the result onto a 40-point spectrum of internal states — not surface behaviours — and tracks trajectory, so the system can tell whether a dog is escalating or settling before the behaviour visibly changes.
• CERT protocols turn that reading into specific, state-aware guidance for the human.
• UCL, the Universal Canine Language, is the output layer that renders the state as speech.
• DAISY, a private per-dog model, adapts to the individual animal over time. It is not shared and not pooled.

The output is not a dashboard label. Instead of “your dog is anxious,” the dog says: “I need you close and calm. What just changed in the room?”

The training corpus is drawn from roughly 120 real breeds — recorded animals, not synthetic data.

The hardware is a harness, and that was a deliberate reversal

Doolittle’s reference sensing design is a harness, not a collar. The company built toward a collar first and abandoned it: a collar rotates as a dog moves and loads the throat, which is the single worst place to take a reading and ask an animal to relax. A harness spreads the load across the chest and keeps the sensor stable and the reading true.

The engine is hardware-independent. The browser beta requires no wearable at all.

Why it matters

In 2025, 2.8 million dogs entered U.S. shelters and rescues, and 320,000 were euthanized — roughly 11% of intake (Shelter Animals Count, 2025 Annual Data Report). Two million dogs were adopted, a 57% adoption rate, up from 55% the year before.

The reasons dogs are surrendered are complex and mostly human: housing, cost and life circumstance outweigh animal-related reasons by roughly three to one, and non-aggressive behaviour issues account for about 7.8% of owner surrenders (Best Friends Animal Society). Doolittle does not claim that misread emotion is the primary driver of shelter intake. It claims something narrower and more defensible: that in the fraction of cases where fear is being read as aggression, the cost of the misreading is absolute, and translation is a tractable problem.

The commercial context is a $158 billion U.S. pet industry in 2025, projected to reach $165 billion in 2026 (American Pet Products Association), with 71 million U.S. households — 53% — now living with a dog. The pet tech products category is forecast to reach $23.5 billion by 2033 (Grand View Research).

What ROSIE is not

ROSIE is not a diagnostic tool. It does not replace a veterinarian, a certified behaviour consultant or a trainer, and Doolittle does not present it as a substitute for professional care. The company is actively recruiting behaviourists, shelter staff, veterinary professionals and open skeptics to test the engine adversarially, and independent validation work is underway with the Spanish sighthound rescue SOS Galgos.

“A system that claims to read emotion should be the most tested thing in the room, not the least. We would rather be corrected in beta than believed too early.” — Danielle A. Franklin

Availability

The ROSIE beta is live now and free at rosie.doolittlecorp.com. Company information, the press kit and licensing enquiries are at doolittlecorporation.comdoolittlecorporation.com. A Science of the Canine Mind, Books 1 and 2 by Danielle A. Franklin, is scheduled for Summer 2026.

Licensing conversations are open with pet wearable manufacturers, veterinary networks, shelter management platforms and companion robotics developers.

About Doolittle Corporation

Doolittle Corporation builds AI systems that give a voice to those who cannot speak for themselves. Its first product, TalkingDOG, is powered by ROSIE — the Real-Time Operating Service Interspecies Element — a patent-pending engine that translates canine emotional states into human language and actionable guidance. The company was incorporated in Delaware in March 2026 and is led by founder and CEO Danielle A. Franklin.