news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Text｜Wang Qin

Editor: Deng Yongyi

Ellie is a golden retriever. Her owner sent a voice message to an app on her phone: "HI Ellie, can you take the remote down for me?"

The app translates human speech into dog language — a mix of long barks and short yelps — in under a second.

Then, the golden retriever went to the coffee table and picked up the remote control!

The user asked the golden retriever Ellie to get the remote control. Image source: Trainiapp Youtube

This AI application is called Traini, and it is an AI dog language translator that can perform cross-species translation between you and your dog: you can understand when your dog wants to eat or when he has separation anxiety, and your dog can also understand what you say, "Give me the remote control" or "Hug me if you love me."

Traini is a pet empathy AI app founded by Chinese serial entrepreneur Sun Linjia in Silicon Valley, USA. It has 200,000 registered iOS users in the cold start. In May this year, the product has been viewed more than one million times.

Traini completed a 10 million yuan angel round of financing in May this year, with FutureX Capital, BlueSea Partners, Tao Foundation, Mint Capital and other investors, including former Meta Engineering Director Liang Yingyi, Caper AI (Instacart subsidiary) Engineering Director Lily Zhang, ORH Partners Partner Jia Peilin, blogger Janie Tanmei, etc. Old shareholders Valkyrie Fund I LP and Fengmin Gong (Palo Alto Networks Cofounder) continued to invest.

Traini not only allows users to understand the dog's barking, but also understand the dog's facial expressions, body behavior and other languages ​​to understand the dog's psychological state.

When training the model, the voice data of dogs of different breeds and regions, as well as multimodal data of dogs' expressions and behaviors, are fed to the AI ​​big model. Since then, the big model can also understand dog language.

Classification of dog behavior signals. Image source: Dog Trainer

Collected 100,000 dog sounds and expressions, with an 80% translation accuracy

The scene of cross-species translation between humans and dogs is not from a movie, nor is it the dog language recognition app that was used to make money. Instead, it is based on facts and logic.

In order for the AI ​​model to learn and understand the dog's language, we must first ask pet behavior experts to translate the meaning of the dog's various expressions, behaviors, and calls and establish a semantic system.

Then, the model is trained with the collected dog data. Currently, Traini has collected the lifelike sounds and expression pictures of more than 100,000 dogs from all over the world, and established a diverse database of pet emotional expressions.

Sun Linjia said that the neural network Dr Traini developed by his team has increased the accuracy of behavior translation to 80%.

There are different ways to train a bark translation model. One is to train the model from scratch using dog voice data; the other is to fine-tune the model using dog voice data based on a pre-trained human voice model.

Regarding the method of pre-training with human voice and then fine-tuning with dog voice, Sun Linjia said that there are already mature and recognized standards for using human voice training models, so you can try to use dog voices as a benchmark. For example, the sound of an angry dog ​​is similar to the sound of an angry person.

In addition, since dogs of different breeds and from different countries have different barking voices, Traini will also use other methods in the training model, such as mixed barking training between dogs of different breeds and voice training of dogs from different countries.

Traini diagnoses symptoms based on the dog's barking. Image source: Trainiapp Youtube

In addition to being able to translate dog language, another major function of Traini is pet medical care - diagnosing whether the pet is sick and what disease it has based on the pet's behavior.

For example, when you enter a dog barking in a lazy voice in Traini's chat box, Traini will quickly reply that this is because the dog is bored.

Traini also looks at the psychological and medical factors behind boredom symptoms in dogs: "This is generally caused by a lack of stimulation in a dog's environment, and the dog is seeking interaction and activity. The barking of boredom in dogs is often repetitive, indicating a lack of physical exercise or mental activity."

Traini also gives medical advice: "Give your dog regular walks, playtime, interactive toys... For more comprehensive medical advice, please consult a veterinarian or pet medicine specialist."

If a dog has separation anxiety, such as a worker being away from home for a long time and the dog is alone at home, depressed, or even has a stress reaction, Traini will identify and diagnose the dog's symptoms and recommend sending the dog to a foster home.

Would you be envious at this moment? When working people have no time to care about their own mental health, Traini is already caring about the mental health of dogs.

In addition, Traini, as its name (Train) suggests, can help users train their dogs.

Many dog ​​owners have misunderstandings when training their dogs, and there are various unscientific training methods, such as intimidation, beating and scolding. Without understanding the dog's meaning, the dog owner's training direction may go in the opposite direction.

In this regard, Traini provides various dog training courses, allowing users to scientifically teach their dogs how to sit, turn around, play with a ball, shake hands, and even do yoga.

Discover pet business opportunities from delivering dog food

In addition to the "dog language translation" function, Traini also takes the commercial route of offline services, connecting the personalized understanding of pets online with offline pet medical care, foster care and other services. For example, users can use dog language translation on Traini to understand that the pet's uncomfortable cry is because it is sick, and then they can place an order on the app with one click and go directly to the nearby pet medical institution.

From 2023 to 2024, AI big models have expanded from the model layer to the application layer, and the same is true in Silicon Valley. "A year ago, we were optimistic about AI applications and made applications, and we were laughed at as idiots (a year ago, Silicon Valley was rolling up big models instead of applications). After a year, making applications has become a consensus," said Sun Linjia, founder of Traini.

Traini is not an application specifically designed for AI, but was born for the pet market. Before ChatGPT came out, Sun Linjia registered Traini in July 2022, targeting the pet market.

He believes that we should first find the application scenario and then use technology according to the scenario, rather than using a hammer to find nails. "No matter in what era or what period of opportunity, the leading edge in technology belongs to only a few companies, such as OpenAI and Google, and is not universally applicable to all entrepreneurs. Every entrepreneur should start from what he is good at and find scenarios in the application layer."

Sun Linjia's entrepreneurial experience in both China and the United States has accumulated a nose for finding application scenarios. He founded the snack brand "Neighbor Shop" and the beauty O2O (Online to Offline) platform in China, and has 12 years of entrepreneurial experience in building digital platforms for life services. After arriving in the United States, he served as CEO of GESOO, a Chinese food delivery platform in the United States, and as the West Coast head of Chowbus, an Asian food delivery platform in the United States. Chowbus has received more than $100 million in financing.

Sun Linjia told Intelligence Emergence that he first discovered business opportunities in the pet market when he was running a food delivery platform.

Around 2018 and 2019, when Sun Linjia was working at Gesoo, a Chinese food delivery platform in the United States, he found that female users on the platform often ordered dog food for their dogs. At that time, Gesoo had hundreds of orders a day, and the number of users who ordered dog food delivery could reach hundreds per month.

At that time, the pet food brand Just Food For Dog had already appeared in the United States, which focused on providing dogs with freshly made dog food, delivered immediately to users' homes, and had a high average order value. This is just like ordering takeout, except that the service targets dogs instead of humans.

In the United States, dogs have the same status as family members in the family. Even if the mother is pregnant, the dog will not be given away. If the dog has cancer, many owners are willing to spend tens of thousands of dollars to treat the dog.

According to statistics from 2022, there are more than 260 billion US dollars of pet consumption in the world, and it continues to grow at an annual rate of 18%. The United States accounts for more than half of the global market share, exceeding 160 billion US dollars.

It was only during the pandemic that there were economic instability factors, and it was not until July 2022 that Sun Linjia actually started to establish Traini.

Sun Linjia believes that the pet market in the United States is huge, but the degree of digitization is not high. He initially wanted to develop a community application for pets. In the middle of this, ChatGPT came out in November 2022. He then wanted to use the technology of big models to translate pet behavior, which was still a blank in the market at that time.

At present, Traini has developed its own small model specifically for pet behavior translation, ensuring that pet behavior data is in the hands of its own company.

The company team is also gradually expanding. Co-founder Jason was a member of the Hunan Province Mathematical Olympiad Youth Training Team, and later studied computer science at Northeastern University in the United States. He has worked at OpenAI and Baidu USA. Growth consultant Yiyi has more than 10 years of experience in product recommendation and data growth. He received a PhD from Brown University and was the head of data growth at CVS and Chewy.

Traini APP can be combined with offline pet services. Image source: Traini

In terms of revenue model, Traini currently has plans for both To B and To C. In terms of To B, Traini provides APIs to pet clinics and pet hardware companies, allowing pet doctors to use Traini to better understand pet behavior, achieve auxiliary medical treatment, and enable pet hardware to interact with pets. Currently, Traini has established cooperation with many hardware companies such as Dreame Technology.

For C, Traini offers pet training courses that users can pay for. Currently, Traini has set up two subscription packages, 12.99 USD/month and 99-129 USD/month. Compared with the offline pet training fee of 300 USD per class in the United States, Sun Linjia believes that the price of the courses in the Traini app will be competitive. In addition, Traini will set up advertising to generate revenue in the future, but currently no advertising is added for the sake of user experience.

In terms of revenue strategy, Sun Linjia’s main goal is not to monetize through To C services in a short period of time. He prefers to provide API interfaces to companies to generate revenue.

In the long run, Sun Linjia believes that AI to Offline (connecting offline services with AI) is the opportunity for AI application. Only when AI is integrated into offline life can it change people's lifestyles. Therefore, Traini not only translates pet language online, but also helps users complete various tasks on the APP, such as searching for information, ordering products, making doctor appointments, buying medicine, etc.

AI applications such as online document office tools mainly provide information. In contrast, offline services can provide effective solutions. "After all, a dog is a real pet. When a pet really has a problem, I need to buy medicine, food, and clothes for it. We can understand the pet's behavior in a personalized way based on the user's description and then match these offline solutions."

In terms of business model, AI to Offline is similar to OTO (Online to Offline) in the mobile Internet era. Currently, the offline service of pet AI is very similar to the previous takeaway delivery startup. "Takeaway is nothing more than us finding restaurants and signing them up on this online platform. Now we use AI to match (dogs' needs with pet offline services), which is more accurate and efficient," said Sun Linjia.

Zhang Qian, founding partner of Skyline Capital, said: "With the rapid development of artificial intelligence technology today, we are particularly optimistic about innovative companies that can deeply cultivate vertical fields and accumulate rich data and professional knowledge. Traini is such a model. It cleverly integrates AI technology into the vertical field of pet economy, not only breaking through the limitations of traditional pet behavior training, but also expanding into the field of pet health with the power of data.

We have always believed that emotional value is an important direction for the application of big models, and it is also a deep need of today's society, families and even individuals. Traini uses its AI technology to not only provide pet owners with more personalized and efficient services, but also deepen the bond between pets and pet owners on an emotional level, allowing pet owners to deeply feel the warmth and care brought by technology."

In the crowded arena of AI applications, in addition to office tools, agents, and social companionship, AI to Offline may be a new track that can truly change people's daily lifestyles, but its future effects still need to be verified by the market.