Deepingsource

Market position

Positions SAAI against POS systems, foot-traffic sensors, conventional single-camera video AI, general-purpose AI, telco and transit-card data and manual counting, arguing that only its approach reads in-space behavior such as dwell, path, entry/capture rate and item pick-up while retaining no identity data. Claimed credentials include 103 patents (66 registered, 37 pending), SOC 2 security controls certification in 2024, and a 'Top Innovator' award at NextRise 2024.

The company frames its distinction as anonymization-first architecture: because identity is erased at capture and raw footage is never retained, footage can be analyzed and linked across cameras without individual opt-in or privacy exposure, and no human review is required. It contrasts SEAL's partial, utility-preserving anonymization with blur and masking, which it says lose detail and reduce model accuracy, and contrasts SAAI's space-specific structure and history with general-purpose AI that gives generic answers from aggregate numbers.

Technology

Identity removal occurs at the moment of video input, so raw footage is not stored and, per the company, anonymized data cannot be re-identified; processing happens in volatile memory. Because identity is erased first, the system can re-link the same subject across camera views using multi-target multi-camera tracking and 3D coordinates mapped onto a floor plan. SEAL's four stated design principles are Secure, Embeddable, Adaptable and Lightweight, with a lightweight runtime able to anonymize live streams in constrained environments. The company states that SEAL-processed data remains usable for vision tasks including action recognition, car detection, semantic segmentation, depth estimation, face landmark detection, person attribute recognition, pose estimation, and gender/age classification. Two public code repositories exist under the DeepingSource GitHub organization, covering noisy adversarial representation learning and medical image obfuscation for distribution.

Go-to-market

Direct enterprise sales through requested consultations and adoption trials on sample footage, supported by case studies and a steady stream of press releases and media coverage. The company also participates in industry events and government programs, including exhibiting its store-operations AI at the Franchise Startup Expo at COEX and selection for Korea's Ministry of Trade, Industry and Energy AI commercialization project. It is an NVIDIA Inception partner.

Operators of brick-and-mortar and unmanned retail spaces — convenience stores, health-and-beauty and franchise retail chains, automotive showrooms and other physical venues — plus organizations that need to de-identify video for AI training and data sharing, which consume the capability through the SEAL SDK. Named customer spaces include Olive Young, 7-Eleven, Don Quijote and MARUHAN.

Geography

Headquartered in Seoul, South Korea (Gangnam-gu; a directory lists 312 Teheran-ro, Vision Tower 14F). Customer references and press coverage are predominantly South Korean, and the company maps its anonymization approach to Korean PIPA, EU GDPR and US CCPA definitions of anonymous or de-identified data.

History

Founded in 2018, the company began with video anonymization — erasing identity as footage enters the system. In August 2021 it obtained an interpretation from Korea's Personal Information Protection Commission regarding anonymous data, and in December 2021 completed a legal opinion on its anonymization technology. Public code repositories were last updated in 2022 and 2023. It received a 'Top Innovator' award at NextRise 2024 and SOC 2 certification in 2024, and subsequently extended from de-identification into spatial analytics and agentic AI for physical spaces, unveiling spatial agentic AI for real-time store checks in April 2026 and presenting SAAI at the Franchise Startup Expo in August 2026.

Risks & controversies

The business depends on the legal treatment of anonymized video; the company cites Korea's PIPA Article 58-2, GDPR Recital 26 and the CCPA, along with a 2021 regulator interpretation and legal opinion, as the basis for its data not being personal data — a position tied to evolving privacy regulation and enforcement.