n7kel Vakhta platform overview

1. Purpose

1.1. n7kel Vakhta (hereinafter, the platform) is a software system for staff video analytics on site. The platform analyzes the video stream from IP cameras that are already installed, records staff-related events together with the supporting evidence, passes violations to responsible persons for review, and produces analytics on how work at the site is organized.

1.2. The platform is used at sites with permanent staff posts: fuel stations, production sites, warehouses, sales floors, cash handling areas, healthcare and office facilities.

1.3. n7kel AI is trained and tested on ordinary 2D images, so the cameras you already have are suitable. 3D cameras are supported too, if they are already on site or the task requires them.

1.4. Recognition is performed by n7kel AI locally, on site equipment. No foreign or cloud recognition services, external APIs or subscriptions are used, and the video stream is not transmitted outside the site.

1.5. n7kel AI is trained further for the client’s tasks — new actions, objects, uniforms, procedures and scenarios — on annotated examples from the site, at the client’s decision.

2. Platform components

2.1. The platform consists of:

• the n7kel on-site unit — a computing device installed at the site. It receives video from cameras, performs recognition, records events with a frame and a video clip, signs them and adds them to a chain of records;
• the server — stores events, violations and POS transactions, performs POS reconciliation, produces analytics, sends notifications and provides an interface for external systems;
• the web interface — the users’ workspace in a browser: dashboard, violations, event log, analytics, checkout, integrity, reference data, settings;
• the independent evidence verification program — allows an exported evidence package to be verified on any computer without contacting the server and without the vendor’s involvement.

2.2. The server is hosted on the client’s hardware (autonomous option) or in a data center in the client’s country (hybrid option). The options are described in the Deployment document.

3. Operating principles

3.1. The platform operates on five principles. They should be taken into account during deployment and when reviewing events.

1. Raw video never leaves the site. Only events, a frame captured at the moment of the event and a short video clip of the event (about 10 seconds: 5 before and 5 after) are transmitted outside the site. In the autonomous option nothing is transmitted outside.
2. Silence over noise. If an action or object is not clearly visible, the result is “undetermined”, not a violation. A false accusation is considered worse than a missed event; the share of false alarms is the first quality indicator and the first indicator on the dashboard.
3. Evidence together with the event. Every event is signed by the on-site unit at the moment it is recorded and added to the chain of records together with fingerprints of the frame and video clip. Retroactive editing is detected by verification.
4. A person makes the decision. A violation is a signal for review by a responsible person. No measures are taken against an employee automatically; a violation can be closed only with a comment, including as a “false alarm”.
5. “No reference” is not a mismatch. If there is nothing to compare a person with (no consent or no reference), the platform draws no conclusions about identity and makes no accusations.

5. Fuel Station Checkout module

5.1. The module compares what the camera above the cashier’s position saw with POS transactions. A violation is opened only when the discrepancy is confirmed by both the video and the POS data.

5.2. The camera records the opening of the cash drawer, payment via the terminal, goods being handed to a customer, a customer’s presence at the counter and who is working at the till. The POS system sends receipts, refunds, voids, no-sale transactions, cashier logins and shifts — via API or as a CSV/XLSX file.

5.3. A violation is not opened if there is no POS data for that moment; if the camera did not register a customer when goods were handed over; if the cashier’s login is not mapped to an employee or the employee has no reference; or if the logged-in employee was the one working at the till. A receipt that arrives late clears the violation automatically.

7. Restrictions and confidentiality

7.1. Any restrictions the client needs are put in place at the client’s request: which cameras, posts, zones and modules are enabled, what is recorded, retention periods, and which users have access and with what rights.

7.2. In every case, all information remains confidential and is not shared with third parties. The only exceptions are those expressly provided for by the contract and the law: transfer of match information to the authorized body when the wanted-list check module is enabled (under an agreement with that body), and processing on the client’s instructions in the hybrid option.

7.3. Visitor recognition is off by default. It is enabled only by the client’s decision, as a separate module, and only on a lawful basis: the person’s consent or an agreement with the authorized government bodies. Employees are identified only with their written consent.

7.4. A responsible member of the client’s staff makes the decision on every signal; no measures are taken against employees automatically.

8. Users and roles

8.1. User permissions are determined by role: organization administrator, occupational safety or security specialist, site manager (assigned sites only), auditor (view only). Accounts are personal, and every action is logged under the user’s name.

8.2. Vendor specialists access the client’s data only at the client’s request, and every such access is recorded in the client’s activity log. In the autonomous option the vendor has no access to the data. See the Security and personal data document for details.