# UniverseAI (UNIVERSE AI INC.): full reference UNIVERSE AI INC. was incorporated in March 2024 and is based in Seoul, Republic of Korea. We develop our own face and palm authentication engine and our own video search and compression software, and we put them into finance, public-sector, transit and retail sites. The customer's cameras and terminals stay as they are; what we build is the software that runs on top of them. Recognition, liveness, indexing and compression are all developed in-house at our corporate R&D centre. Site: https://univs.ai/en/ · Contact: team@univs.ai · Developer docs: https://docs.univsgate.com No prices are published on this site. Quotations are prepared once the enrolled population, the deployment mode (own servers or cloud) and the target deployment date are known. Customer names are deliberately kept off the site. --- ## GATE, biometric authentication platform https://univs.ai/en/gate GATE is authentication software installed on the customer's own servers. Face and palm engines run inside it, and it faces the site systems through one documented REST API, so surveillance, attendance, payment and access all call the same server. Because that API stays the same when an engine is added, changing the authentication method becomes a configuration change rather than an integration rebuild. - **Two modes in one server.** A 1:N search finds who a person is within an enrolled population; a 1:1 check confirms that this person owns this account. The two demand different server resources, so the configuration is drafted from the dominant mode, the enrolled population and the daily volume. - **RGB passive liveness.** Ordinary colour cameras, no infrared or other special hardware, and the user does not blink or turn their head. It screens the common spoof attempts: printed photos, video replayed on a screen, and masks. Field performance tracks camera quality and lighting, so thresholds are tuned to the installation environment. - **Face and palm, dual modal.** A palm gives a second route where the face is blocked (masks and hats, backlight, changed appearance), or a second factor for high-value transactions. It requires that the terminal camera can capture it, so capture distance, field of view and lighting are checked first. - **One REST API.** The customer's team or SI partner builds from the documentation, and an issued key allows calls in a test environment first. Boundary: terminals and site systems belong to the customer. GATE's scope runs to the spoof check and the match; payment approval, door control and membership handling are picked up by the customer's systems, and the integration work is carried out by the customer or their SI partner from our specification. Deployment and data: on-premises installation on the customer's own servers is available, and it can also run in cloud and container environments. Under an on-premises configuration, biometric data does not leave the customer environment, and what is stored is feature data for matching rather than the original image. Measured figures: 99.97% accuracy under KISA certification testing, a 400M+ face database scale, and a 1:N search response under one second on the server side. The 99.97% figure was measured under accredited test conditions; field numbers move with camera specification, lighting and enrolment image quality, so verification in the real environment is recommended first. --- ## VCA, forensic video search and analytics https://univs.ai/en/vca VCA is a video-analytics brain that attaches to the ordinary cameras already installed. Nothing is swapped out for AI cameras; it takes the video already being recorded, finds people and vehicles in it and organises them, so a person can later be traced back from a single photo or a few conditions instead of rewinding days of footage by eye. Search axes: face (a 1:N search from one photo); body, meaning clothing and build, which finds the person when no usable face was captured and is the axis most used in the field; colour attributes; bags and belongings, because clothing changes and the bag often does not; RedFace, a companion network answering who this person moved with; and RedMap, a GIS movement trail plotting where the person passed through. Search combines filter conditions shown as chips, producing a candidate list that can be played back on the spot. It is not queried in free-form speech, and it does not make the identification: it narrows the candidates, and the customer's officer makes the final determination. Cameras are not replaced. VCA connects over ONVIF and RTSP, so brand does not matter and there is no replacement cost, and the existing VMS is kept in place and runs alongside. On-premises and cloud are both supported. Ordinary 2MP CCTV is the minimum image quality; very old, low-resolution or badly angled cameras lose accuracy, so the minimum standard is confirmed beforehand and the usable cameras identified. Track record: 20 deployments across 5 countries. --- ## AI BOX, edge compression and indexing https://univs.ai/en/aibox Retention days equal storage capacity divided by daily footage volume, so moving to 4K multiplies the volume and shrinks retention by the same factor. AI BOX compresses the video next to the camera, shrinking that denominator, and builds a face index while it does. The cameras stay and the existing VMS stays. At 4K 16Mbps on unchanged storage, footage left as-is lasts about 7.5 days; a codec upgrade roughly doubles that to about 15 days, still short of the usual 30-day operating policy; AI BOX, at about 85% reduction, recovers about 50 days. The differentiator is not how much is reduced but whether the policy line is cleared. Exactly one new block goes in, between the cameras and the storage. Because compression finishes beside the camera, no central server room, cooling or extra infrastructure is required, and sites without a server room are the typical customers. Processing is local, so when the link drops, compression and recording continue. Configuration: 4 channels per box, scaling by one box per 4 channels with no server expansion; ONVIF and RTSP alongside the existing VMS; 4K (UHD), 1080p, H.264, H.265 and MJPEG supported; about 85% reduction, measured on site and varying with scene composition; measured 4K bitrate about 2.1Mbps. A standard line on a low-power edge processor handles 4 channels at 2 frames per second, and a higher-tier line with an NPU covers real-time processing and 4 to 8 channels. On image quality, the accurate phrasing is a level at which degradation is hard to perceive, best judged by viewing the original and the compressed stream side by side in a demo. --- ## POS, face and palm payment terminal https://univs.ai/en/pos A payment terminal that takes face or palm and also handles IC, NFC and receipt printing in a single device. Enrolment happens once; after that, no wallet and no phone. Customers who do not enrol keep using their card, so replacing the terminal does not split customers into two groups. The flow: a one-time enrolment links a face or palm to a payment method. At every later payment the customer stands at the terminal, a liveness check screens photo and video spoof attempts, and a 1:N match runs against the enrolled data. Our segment ends there. Authorisation and settlement happen on the customer's PG and payment network, and a built-in thermal printer issues the receipt. Configuration: a 7-inch IPS touchscreen, an RGB 5MP and NIR 5MP dual camera, face and palm (vein) biometrics, passive liveness requiring no extra user action, IC, NFC and magnetic stripe modules built in, API integration with the existing POS system, and on-device inference running alongside the payment app. EMV and PCI PTS certification is planned, and the completion date will be notified separately once fixed. Where it fits: places the same people come back to, such as staff canteens and shops settling against payroll deduction, sites that tie building access and payment to one face, and membership stores. Where it does not fit: stores that see only one-time visitors, because the enrolment step makes the counter slower rather than faster. In that case conventional payment is the better answer, and we say so. --- ## Use cases https://univs.ai/en/solutions/ **City control centres** (https://univs.ai/en/solutions/city-control/). Operators scrub through days of footage by eye, and after a 4K upgrade that footage is sometimes already gone. Both are solved without replacing a camera. **Financial authentication** (https://univs.ai/en/solutions/finance/). Remote onboarding rests on a submitted image, and deepfakes have made forged documents and replayed video easy. The engine is installed inside the institution's own servers, so spoof detection and matching finish inside its perimeter. **Access control** (https://univs.ai/en/solutions/access/). An entry log records that a card passed through, not that a person did. The image sent by the entry terminal is matched on the server to verify the person themselves; opening the door remains the job of the access system already in place. **Campus and workplace** (https://univs.ai/en/solutions/campus/). Calling names verifies a voice and tapping a badge verifies an object. At the classroom, office and canteen door the person is verified by face, and the result is returned to the academic, HR and canteen systems. **Field and industrial** (https://univs.ai/en/solutions/field/). No server room, a poor or temporary link, and the site moves. One box sits next to the camera and cuts the volume sent: on a wireless link a smaller monthly bill, on fibre more retention days on the same storage. --- ## Certifications and record KISA face recognition accuracy testing (99.97% under certification conditions); iBeta liveness presentation-attack testing; GS (Good Software) Grade 1 for 'UNIVS AI Guard'; ISO 9001 and ISO 14001; public procurement registration in Korea; a corporate R&D centre where engine development is done in-house. Core R&D began in 2021, the VCA beta and first pilot followed in 2022, and the company was incorporated in March 2024. --- ## Frequently asked questions https://univs.ai/en/#faq **What sort of company is UniverseAI?** Incorporated March 2024, based in Seoul, Republic of Korea. We develop our own face and palm authentication engine and our own video search and compression software. Your cameras and terminals stay as they are. **What is GATE?** A biometric authentication platform installed on your own servers. Face and palm engines run inside it, and it faces your site systems through a single REST API. It performs 1:N matching, 1:1 verification and passive liveness on ordinary colour cameras. Accuracy is 99.97% under KISA certification testing. **What is VCA?** Video search and analytics software layered onto the CCTV you already have. It connects over ONVIF and RTSP, so camera brand does not matter and your existing VMS stays in place. You combine conditions to narrow the candidates, and your officer makes the final determination. **What is AI BOX?** An edge appliance installed next to the camera. It compresses footage by about 85% on the spot, so retention days survive a move to 4K, and it builds a face index while it compresses. One box covers four channels and no central GPU server is required. **Where is it used?** Smart city control rooms, campus and workplace attendance and canteens, face payment and identity checks at the POS, site access and attendance, financial remote authentication, and field and industrial sites with no server room. **Where is the data stored?** Under an on-premises configuration the data is stored inside your own servers and biometric data does not leave your environment. What is stored is feature data for matching rather than the original photograph. Cloud and container deployments are also available. --- Questions, demo requests and integration scope reviews: team@univs.ai Korean versions of every page live at the same path without the /en/ prefix.