Questions about OCTOps answered

OCTOps is an AI visual monitoring platform that connects to your existing cameras and runs multiple AI models simultaneously — monitoring quality, safety, inventory, labour, and equipment in real time.

Unlike traditional CCTV, which records footage without understanding it, OCTOps analyses every camera feed continuously and surfaces insights, alerts, and events to a unified dashboard your team can access from anywhere.

The platform runs on an on-site edge machine — meaning all video is processed locally on your premises. Only results, alerts, and event-triggered clips leave your facility.

No. OCTOps is designed to work with the cameras you already have. The platform connects to existing IP cameras, CCTV systems, and industrial cameras via your network — no replacement, no new cabling, no hardware lock-in.

The only hardware requirement is an RTSP stream from your cameras, which is a standard setting on most IP cameras. If yours don’t currently have this enabled, our team can assist with configuration during onboarding.

OCTOps has been deployed across multiple facilities without a single camera being replaced.

OCTOps is proven across manufacturing, logistics and warehousing, construction, and commercial facilities. Any operation that relies on visual oversight – quality control lines, inventory management, equipment monitoring, or workplace safety is a strong fit. We also support custom model development for niche industry requirements.

The edge machine is a compact computing unit installed on your site that runs all AI models locally. It connects to your cameras via your network and processes video in real time — without sending raw footage to the cloud.

We use an edge-first architecture for two reasons:

• Data privacy: Your operational video never leaves your facility. Only aggregated results and event-triggered clips are sent to the cloud.
• Performance: Processing video locally eliminates latency. AI analysis happens in real time, not after a round-trip to a remote server.

A single edge machine can handle 10 or more cameras simultaneously, with multiple AI models running at the same time.

Multiple capabilities can run simultaneously on the same camera feed. OCTOps has seven built-in capabilities: Labour analytics, Inventory management, Equipment panel monitoring, Visual quality inspection, Safety management, AI system integration, and AI model enhancement.

For example, a single camera on a production floor can simultaneously monitor worker PPE compliance, detect equipment panel anomalies, and track zone activity — without requiring separate hardware or separate vendor contracts for each.

Yes. OCTOps is designed to feed into your existing workflows rather than replace them. Integration options include:

• ERP systems: Inventory counts and labour data feed directly into your ERP without manual data entry.
• PLC integration: For visual inspection, the system can communicate directly with your PLC to halt a machine the moment a defect is detected — before the faulty unit is sealed or dispatched.
• Production management systems: Man-hour data matched to PLC output can be fed into your production management platform for cross-site performance comparison.

Custom API connections are also available for systems not covered by standard integrations.

The unified dashboard provides live visibility into all active AI models and camera feeds across your operation. Depending on which capabilities you have deployed, it surfaces:

• Real-time alerts and event notifications with timestamps
• Event-triggered video clips for immediate review
• Inventory counts and stock level updates
• Labour analytics and man-hour data matched to production output
• Equipment status and anomaly records
• Safety compliance status by zone and shift
• AI model accuracy metrics and improvement cycle progress

The dashboard is accessible via web browser and mobile — no dedicated hardware or on-site presence required.

Yes. The OCTOps AI Integration capability is designed to connect existing AI tools, machine learning models, and automated systems into the platform alongside the built-in capabilities. This eliminates fragmented deployments and separate vendor dashboards.

Data from all connected models — including third-party ones — is pooled in the shared platform, enabling retraining, anomaly correlation, and unified visibility from a single dashboard. New AI capabilities can be added over time without disrupting existing deployments.

Most deployments are live within 7–14 days of the initial assessment. The process has five stages:

• Week 1: Operational assessment — camera inventory, use case prioritisation, integration mapping.
• Weeks 1–2: Edge machine installation and camera connection.
• Day 7–14: Go-live with baseline AI models. Dashboard active.
• Weeks 2–6: Real-world data collection for the first retraining cycle.
• Month 2 onwards: First improved models redeployed. Continuous improvement from here.

OCTOps is live in days, not months. The baseline models work from day one — and improve from day two

The requirements are minimal. You need:

• Existing IP cameras or CCTV with accessible RTSP streams
• A network connection to the camera system on-site
• A power outlet and rack or shelf space for the edge machine
• Standard broadband internet for cloud dashboard access
• Agreement on priority use cases before installation begins

You do NOT need new cameras, new cabling, new network infrastructure, a dedicated server, an in-house AI team, or any specific camera brand or ecosystem.

Minimal disruption. The edge machine connects to your existing cameras via your network — there is no new cabling required and no changes to your camera infrastructure.

Installation is typically completed in a day or two, and does not require production to stop. Once live, your team’s only involvement is reviewing alerts in the dashboard — the AI monitoring runs continuously in the background without requiring operator attention.

Yes. Camera and zone assignment is fully configurable. You can:

• Assign specific AI models to individual cameras or camera groups
• Define zones within a camera view with different rules (e.g. PPE required in Zone A, restricted entry in Zone B)
• Set different alert thresholds per zone
• Configure alert routing to specific team members by site, zone, or event type
• Schedule models to run during production hours only, or 24/7 for safety applications

Yes, and this is the recommended approach. We work with your operations team during the assessment to identify the highest-priority use case for your site — often visual inspection, safety monitoring, or inventory counting — and deploy that first.

Additional capabilities can be added over time on the same cameras and the same edge machine, without disrupting what’s already running. The platform is designed to scale with your operational needs rather than require a full deployment upfront.

• Day 1–7: Baseline AI models running. Dashboard live. Team trained on alert review workflow. First detections surfacing.
• Week 2–4: Alert thresholds refined based on early results. Data collection building toward first retraining. ERP and system integrations verified. Weekly review with your local OCTOps partner.
• Month 2: First improved models redeployed. Accuracy improvement report shared. False positive rate measurably lower. Second data collection cycle begins.
• Month 3+: Continuous improvement cycle running. Expansion to additional cameras or capabilities. Quarterly performance reviews.

No. OCTOps is built so your operations team can use it without any AI expertise. Your team interacts only with the dashboard — reviewing alerts, monitoring live feeds, and adjusting notification preferences.

All AI model management — annotation, retraining, redeployment — is handled by the OCTOps team on your behalf as part of the ongoing partnership. You don’t need data scientists, machine learning engineers, or specialist IT staff to get value from the platform.

No. This is a fundamental design principle of OCTOps, not a policy choice. All video processing happens on the edge machine installed on your site. Raw footage is never transmitted off-site by default.

What does leave your facility:

• Alert metadata (event type, timestamp, camera ID, confidence score)
• Event-triggered video clips — short recordings before and after a detected event, sent to cloud storage for dashboard review
• Aggregated analytics data for the dashboard

All of this goes only to the cloud destination you configure — your own cloud storage or the OCTOps secure cloud.

Video is processed on-site by the edge machine. Raw footage never leaves your facility — by architecture, not policy.

You do. All operational data, video footage, and event recordings generated on your site belong to you. OCTOps does not use your operational data to train models for other clients.

When your data is used for retraining your own models, it is used only to improve the models deployed on your specific site. Your data is never pooled with other clients’ data for general model training.

OCTOps monitors operational activity and events — not individual workers. The system detects and flags specific conditions: a PPE violation, an inventory discrepancy, a machine anomaly, a restricted zone breach. It does not track named individuals, generate employee profiles, or record personal behaviour data.

For example, the safety monitoring capability detects that a worker in Zone B is not wearing a safety vest — it alerts a supervisor. It does not identify who that worker is, log their movements, or create a record tied to an individual’s name.

We recommend discussing the deployment with your team and HR prior to go-live — transparency about what the system monitors and what it does not is the most effective approach.

Yes. Data retention policies for event clips and alert records are configurable. You set the retention period based on your operational and compliance requirements — whether that’s 30 days, 90 days, or longer for audit purposes.

Access controls are also configurable — you determine which team members have access to which clips, dashboards, and historical records.

Since raw video never leaves your site, the footage on your edge machine and camera system remains entirely under your control at all times — including after the contract ends.

Event clips and alert records stored in the cloud are exportable and can be deleted on request. OCTOps does not retain your operational data after the relationship ends. There is no lock-in to a hardware ecosystem — your cameras continue to operate as they always did, and the edge machine is removed.

Baseline models — deployed at go-live — are trained on broad datasets from similar operational environments. Their accuracy varies by use case and environment, but typically starts in the range of 60–75%.

Through the continuous improvement cycle, models are retrained on data from your specific site — your lighting, your equipment, your workflows. By cycle 3 or 4 of real-world retraining, accuracy commonly reaches 85–93%+.

False positives decrease significantly with each cycle as the model learns to distinguish genuine anomalies from normal variation in your specific environment.

One client’s visual inspection model reached 93%+ accuracy by cycle 4 and has operated reliably for 5+ years since.

OCTOps uses a six-step continuous improvement cycle: Deploy → Collect → Annotate → Retrain → Redeploy → Repeat.

After the baseline models go live, the edge machine continuously collects footage from your specific environment. OCTOps specialists annotate this data — labelling what the AI should have detected, what it correctly identified, and what it missed. The annotated dataset is used to retrain your models, which are then redeployed with measurably better accuracy.

This cycle repeats indefinitely. Each iteration produces a model that knows your facility better — your equipment types, your lighting conditions, your production rhythms, your defect patterns.

OCTOps visual inspection can detect a wide range of defect types including:

• Surface defects: cracks, scratches, dents, discolouration
• Foreign objects or contamination
• Dimensional variations and incorrect measurements
• Printing and labelling errors
• Incorrect component placement or mounting
• Packaging defects: missing slips, misalignment, incorrect sealing
• Installation inconsistencies (e.g. spacing errors in repetitive assembly)

Specific defect types are trained into the model during the initial deployment based on your production environment. New defect types can be added through the retraining cycle as your product range or quality requirements evolve.

Yes. The equipment monitoring capability uses AI-OCR to read both digital and analogue displays — including older equipment that predates networked monitoring and has no digital output, API, or integration capability.

This means OCTOps can monitor equipment panels on legacy machinery without requiring any hardware modification, software integration, or direct connection to the equipment itself. The camera sees the display; the AI reads it.

When an abnormal reading is detected, the system automatically records video from the minutes before and after the event — providing a complete record for root cause analysis without anyone needing to be on-site.

OCTOps inventory counting uses AI image recognition from fixed-position cameras to detect and count items visually — without barcodes, RFID tags, or any contact with the items themselves.

This approach is particularly effective for large, bulky materials — pallets, pipes, scaffolding boards, and other items that are difficult to tag or scan individually. The AI is trained to recognise the specific item types on your site and count them as they arrive, move, or leave the camera’s field of view.

Counts are updated in real time and can feed directly into your ERP or warehouse management system via integration — replacing the periodic manual stocktake entirely.

Yes. OCTOps is built on a platform that supports custom model development for operational requirements outside the seven standard capabilities. If you have a specific detection task, process monitoring requirement, or quality check that doesn’t map to the standard suite, the OCTOps team can scope and develop a custom model for your environment.

Custom models run on the same edge machine, feed into the same dashboard, and go through the same continuous improvement cycle as the built-in capabilities. Contact your local partner to discuss your specific requirements.

Ongoing support is part of the OCTOps partnership — not a separate contract. It includes:

• Continuous model management: Every retraining cycle is managed by the OCTOps team — annotation, retraining, and redeployment on your behalf.
• Integration maintenance: As your ERP, production systems, or camera infrastructure changes, our team updates the connections to keep data flowing into your workflows.
• Regular reviews: Weekly reviews in the first month, moving to quarterly performance reviews once the system is settled.
• Local partner support: Your deployment is managed by a regional partner in your timezone — not a remote vendor support queue.

OCTOps is deployed and supported by local implementation partners — not a centralised remote team. Your partner is based in your region, operates in your timezone, and is familiar with your local operational context.

• Australia, New Zealand, Southeast Asia: Mitrais
• Japan: CAC Holdings (platform developer)
• China: CAC Shanghai
• India: ISL

This model means your assessment, installation, ongoing support, and model management are all handled by a team you can reach during your business hours.

If the edge machine goes offline, AI monitoring pauses for that site — your cameras continue recording as normal, but AI analysis stops until the machine is restored. Your local partner will be notified and will work to restore the system promptly.

Event clips and alert records generated before the outage remain in cloud storage and are not affected. The dashboard will indicate that the site is offline so your team is aware.

For business-critical deployments, redundancy options can be discussed with your local partner during the assessment phase.

This is one of the most common concerns we hear — and it’s a fair one. Most AI monitoring tools fail for one or more of the following reasons:

• Generic models that don’t improve: Off-the-shelf AI is trained on general data. Without a mechanism to retrain on your specific environment, accuracy plateaus and alert fatigue builds until the system is abandoned. OCTOps’s continuous improvement cycle directly addresses this.
• No local support after go-live: Many vendors deploy and disappear. OCTOps is supported by a regional partner in your timezone who manages model retraining, integration maintenance, and ongoing performance reviews.
• Hardware lock-in: Some platforms require proprietary cameras or edge hardware. OCTOps works with what you already have — no sunk cost on hardware if the platform doesn’t perform.
• Complexity: Platforms that require internal data science teams to operate are not sustainable for most operations teams. OCTOps is designed for operations managers, not AI specialists.

OCTOps has deployments that have been running continuously for 5+ years. The difference is a platform that improves, support that stays, and hardware you already own.