Verify security checkpoint compliance.
Detect, in real time, whether visitors at controlled access points are actually being reviewed by your team — according to your protocol, frame by frame.
Compliance is asserted in documents, not verified in operations.
At controlled access points — government buildings, restricted zones, public-safety perimeters, manufacturing entries — protocol says visitors are reviewed by a guard before passing. The reality is that compliance is hard to measure. Cameras are everywhere, but the operator who watches them cannot watch hundreds of feeds at once. Audits happen retrospectively, by a human reviewing footage hours after the fact.
The result: protocols drift. Compliance is asserted in documents and not verified in operations. When an incident happens, the after-action investigation discovers gaps in adherence that a real-time system would have surfaced as they happened.
One Workflow answers one question, frame by frame.
Vision is designed for this kind of structured, real-time observation. The Workflow you configure on the platform answers a single question — did the protocol happen? — frame by frame, person by person, with persistent identity across the camera grid.
The configuration uses three layers of the Vision architecture together. The real-time pipeline detects every person entering the scene and tracks them with persistent identity from arrival to exit. The VLM is invoked at two specific moments: once when the person appears (to classify their role — guard or visitor), and once when they leave (to confirm whether the protocol was actually executed during their time in the scene). The event stream publishes a structured event for each visitor with their identity, role, the answer to the protocol question, and the timestamp — flowing into your SOC or audit system through OpenTelemetry traces and the connectors you already use.
No retraining, no labeling pipeline. The protocol changes? You change the prompt. The visual convention changes (different vest color, different uniform, different gesture)? You change the prompt. The pipeline is the same.
Two Workflows in plain language.
// Hypothetical illustration
Take a security checkpoint at an enterprise site. Guards wear orange vests; visitors arrive without vests. Protocol: every visitor must be reviewed by a guard before continuing through the checkpoint. The goal is a real-time event stream that confirms compliance for every visitor.
Two Workflows configured in plain language are enough.
- WHEN
- A new entity is detected (class: person)
- WHERE
- Camera 04 — Main Entry, anywhere in frame
- ANALYZE
- Use VLM Level A on a crop of the person. Prompt: "Is this person a guard wearing an orange vest, or a visitor without a vest?"
- RESULT
- Each person is tagged with their role, once, the moment they appear. The role travels with the entity for the rest of its time in the scene.
The first Workflow runs the moment a person appears in the camera. It uses VLM Level A— a focused look at a single crop — to classify the person's role. The role is attached to the tracked entity, so subsequent stages of the pipeline see the role in every event about that person.
- WHEN
- A tracked entity leaves the scene (class: person, role: visitor, duration > 60 seconds)
- WHERE
- Anywhere in the camera grid
- ANALYZE
- Use VLM Level C on five crops sampled across the entity's time on scene. Prompt: "Was this person reviewed by a guard? At what moment, and by whom?"
- RESULT
- A structured event is published with the visitor's identity, the answer, the timestamp, and a reference to the guard who reviewed them.
The second Workflow runs when a tracked visitor leaves the scene. It uses VLM Level C — a temporal sequence of crops sampled across their time in the scene — to answer the protocol question: was this person reviewed by a guard, when, and by whom. The structured event flowing out of this Workflow is the per-visitor compliance record your audit system needs.
What is striking about this configuration is not what the platform does — that is composability and zero-shot reasoning, working as designed — but what is not required. No model retraining. No annotation pipeline. No data scientist on call. Two prompts and two trigger configurations. If the convention shifts (guards now wear green vests, or the protocol changes to require a documented signature instead of a visual review), only the prompts change. The pipeline stays the same.
A real-time compliance event stream.
A real-time compliance event stream flows into the systems your operations team already uses — SOC dashboards, ticketing, incident management, audit pipelines. The four indicators a typical deployment of this Workflow is designed to support:
- 01Compliance rate
The share of visitors confirmed reviewed during their time in the scene.
- 02Time-to-review
How long elapses between a visitor's arrival and the moment they are actually reviewed.
- 03Drift signal
When compliance rate drops in a specific shift, area, or guard, surface it in real time before it becomes an audit finding.
- 04Audit completeness
Every visitor produces a structured event with its evidence frames; routine compliance audits do not need retroactive forensics.
Specific figures depend on the deployment, the protocol, and the camera grid. They are best discussed against your concrete site on a call.
Adjacent solutions.
Track perimeter access across cameras
Follow a person or vehicle across multiple cameras with persistent identity. Detect entries to restricted zones, lingering, and unauthorized cross-camera movement.
Enforce safety and quality on the production floor
Detect missing PPE, unsafe poses, and quality defects in real time. Route alerts to your operations stack with the evidence frame attached.
Bring your camera grid. We'll walk you through it.
We work with enterprise teams running real-time vision on their own infrastructure. A short call is enough to see if Alquimia Vision is the right fit for your case.
Get in touch