‘Deeping Source’, “Enhance store operations with AI data anonymization technology”

- Deeping Source is an AI and data security company that has developed a 
  proprietary AI-based technology for anonymizing video data, allowing 
  businesses  to use video data safely without privacy concerns.
- They have also created a Store Caring Solution that monitors stores in real-time, 
  providing insights on store conditions, inventory levels, and areas for 
  improvement to enhance operational efficiency.
- Deeping Source has deployed its solutions in various retail locations and is 
  expanding its footprint in Japan through partnerships and the establishment of 
  a  Japanese subsidiary.


‘Deeping Source’ is an AI and data security company that has developed a proprietary AI-based technology for anonymizing video data. The company's unique anonymization solution de-identifies personal information in video footage while preserving the data's intrinsic value, enabling businesses to leverage video data safely and without privacy concerns.

Building on this core technology, Deeping Source has also developed a ‘Store Caring Solution.’ This AI-powered system monitors stores in real time, providing insights on store conditions, inventory levels, and areas for improvement, thereby contributing to enhanced operational efficiency.

Deeping Source has deployed its solutions in a variety of offline retail locations, including convenience stores, beauty shops, and department stores, as well as large-scale venues like COEX and Lotte World in Korea. The company is also expanding its footprint in Japan through partnerships with KDDI and Lawson and plans to establish a Japanese subsidiary this August.

We sat down with Pete Tae-hoon Kim, CEO of Deeping Source, to discuss the company's data anonymization technology and its innovative Store Caring Solution.

Pete Tae-hoon Kim, CEO of Deeping Source / source=IT dongA

Securing the World's Data: The Deeping Source Mission

Could you please start by introducing yourself?

I'm Pete Tae-hoon Kim, CEO of Deeping Source. I have been developing video recognition and analysis solutions for over 25 years. In 2006, I founded the startup ‘Olaworks’, where we developed solutions for recognizing faces and objects using smartphone cameras, right as the smartphone era was dawning. Olaworks was acquired by Intel in 2012, and I continued my work there.

As the importance of privacy and data security came into the spotlight, I realized that developing a solution that could securely protect and analyze data in the AI field would be highly competitive. That conviction led me to leave Intel and found Deeping Source.

What kind of company is Deeping Source?

Deeping Source was founded in 2018 as a specialized AI and data security company. Under the philosophy of 'Making all the world's data secure,' we provide data anonymization technology and our Store Caring Solution.

The company name was inspired by a snack we were eating during a brainstorming session. My co-founder and I were trying to come up with a name while eating nachos with dipping sauce, and the phrase just stood out. So, we decided on Deeping Source. However, the spelling is different. 'Deeping' is inspired by 'deep' from deep learning, and 'Source' signifies source code, origin, or foundation. It embodies our focus on data as the source of AI.

AI-Based Video Anonymization and the Store Caring Solution

What specific technologies and solutions has Deeping Source developed?

We provide our Store Caring Solution, which is built upon our proprietary AI-based video data anonymization technology.

Let me start with the anonymization technology. Data anonymization is one of the most powerful methods of data de-identification, but it is also one of the most difficult to develop. It took us three years just to develop the core technology.

Step1: Anonymize the entire video to completely remove personally identifiable information. Step2: Performing AI video analysis on anonymized data / source=Deeping Source

While video footage doesn't contain explicit information like names or phone numbers, it does show faces, which are considered personal information. Collecting or using them without consent is illegal. We apply noise to these faces to make them unidentifiable. Crucially, we don't just erase the faces; the underlying data remains intact. In other words, we eliminate personal identification while preserving the data's value. This allows AI to fully recognize and analyze a person's location, movement patterns, and posture.

Other de-identification techniques include blurring or deleting facial regions. The problem is that an AI trained on such data may fail to recognize or malfunction when presented with actual, clear data. Our anonymization technology removes personal information while allowing for the safe use of clear, high-value data.

There's also a de-identification technique that replaces a person's face with another. While this preserves data value, it has limitations in real-world applications because it's difficult to prove that the face has actually been replaced. Since our method applies noise to the original data, it's immediately and visually verifiable that the de-identification process has occurred.

Could you also tell us about the Store Caring Solution?

We developed our anonymization technology about seven years ago. At the time, AI was just emerging, and its applications were limited. So, we decided to create a solution that could use this technology ourselves.

We looked for fields where AI adoption was hindered by privacy and data security issues and identified CCTV as a perfect fit. Video from CCTV is rich with personal data and can't be used freely. While security-purpose CCTV installations have broader legal allowances, cameras in retail stores used for customer monitoring and analysis require all personal information to be removed. As a result, their potential was not being fully utilized.

We saw an opportunity to solve this problem and launched a video analysis solution in 2021. It analyzes customer behavior and movement within stores based on de-identified video data. We provided analytics on visitor counts, gender and age demographics, and the movement patterns of specific customer segments, detecting only the minimal characteristics requested by the client.

Store Caring Solution provides owner and employees with real-time information on store conditions / source=Deeping Source

In response to customer needs, we added more AI-based services and launched a store optimization solution early last year. This solution provided reports on operational efficiency, inventory management, and customer service improvements based on in-store data analysis. We further enhanced and expanded its capabilities, upgrading it late last year to the current Store Caring Solution, which provides real-time optimization guidance.

With the Store Caring Solution, owners can check their store's status without being physically present and receive immediate alerts for issues requiring action. It informs employees about tasks that need immediate attention, such as cleaning up spills, restocking empty shelves, or managing low inventory. It also provides management and improvement recommendations to franchise headquarters.

The AI can also find solutions independently when given a goal. For example, if a store frequently runs out of a certain product and the owner asks for a solution, the AI will analyze customer visit and purchase data, compare it with other franchise locations, and suggest a plan to gradually increase order volumes.

Essentially, the Store Caring Solution acts as a middle manager between the store owner and employees, and between the owner and the franchise headquarters. It allows them to experience what it's like to run a store efficiently in partnership with AI.

It seems the data anonymization technology could be applied to many fields beyond store management.

That's right. We have also applied our anonymization technology to safety-related solutions. For instance, a solution that detects when a factory worker runs in a hazardous area or exhibits unusual behavior. It's also been implemented in systems that monitor infrastructure for dents, cracks, or damage. We even built a solution for outdoor advertising that uses nearby CCTV to count how many passersby look at an ad.

We soon plan to test a system that identifies the characteristics of a specific pedestrian, predicts their path, and plays a customized ad on a digital billboard as they pass by. Additionally, we are receiving numerous inquiries from the autonomous driving sector, which is looking to comply with personal information protection laws.

Accelerating Japan Entry, Subsidiary to be Established in August

Could you update us on your business progress so far?

In Korea, our solutions have been deployed in numerous offline retail stores—including convenience stores, beauty shops, and department stores—and large venues like COEX and Lotte World. We are collaborating with several well-known corporations. Each client has different needs, so we provide tailored solutions. The feedback has been positive, particularly regarding the ability to increase operational efficiency.

Recently, our business in Japan has been accelerating, so we are placing a strong emphasis on the Japanese market this year. We are collaborating with retail and logistics companies, centered around KDDI and Lawson. We are also working with a top-tier local Japanese company. As our business continues to expand, we plan to establish a Japanese subsidiary this August to aggressively scale our operations.

You are currently receiving support from the ‘SeoulTech’ Global Corporate Collaboration Program. What kind of support have you received?

We are participating in the Global Corporate Collaboration Program with Amazon Web Services (AWS), which has given us opportunities to join various programs, including networking events and solution training. We also participated in a startup event in Japan this past July.

Furthermore, this program has prompted us to prepare for listing our data anonymization solution on the AWS Marketplace. The AWS Marketplace is a digital catalog where customers can easily find, buy, and manage solutions from AWS partners. We expect this will make it much easier for companies already using AWS to adopt our solution.

Finally, what are your future plans and goals?

We are in an era of rapid AI expansion, so our immediate goal is to quickly capture the market with our data anonymization technology. We will continue to focus on product stabilization and global expansion.

Our long-term vision is to automate the offline world, just as the digital world has been. I believe automation will extend to many more fields. Tasks that are currently handled by people will be automated by robots. Therefore, our long-term goal is to expand into solutions that automate the physical world. Through this, we want to create an environment where everyone who works in offline settings can do their jobs more easily.

By Man-hyuk Han (mh@itdonga.com)

* This article was written with support from Seoul National University of Science and Technology.