Software that watches camera feeds so people only look at the moments that matter.
- Intrusion and perimeter detection after hours
- Loitering and tailgating at doors and loading docks
- PPE checks such as hard hats and high-visibility vests
- Occupancy and crowd counts for safety limits
- License-plate reading at gates, only where the law and your policy allow it
How we build it
We run detection models on an on-site edge server or on the cameras themselves, reading the streams you already have, and send events, not raw video, to your dashboard, phone or monitoring software.
Guardrail Cameras sit on their own segmented network, footage stays in your control, and face recognition is off unless consent and local law are settled first.
A second look at every alarm, so monitoring centers dispatch less often to false alarms and respond faster to real ones.
- Checking a door or motion alarm against the camera at that moment
- Telling a person from an animal, a shadow or blowing debris
- Sending the monitoring center a short verified clip with the alarm
- Escalation rules by site, time of day and event type
How we build it
A model reviews the camera clip that goes with each alarm signal and attaches its finding to the event, in your alarm panel software or at your monitoring center, using the integrations those platforms offer.
Guardrail The AI only adds evidence; a trained operator, not the model, decides to dispatch, and every decision is logged.
Integration and monitoring for machines that clean, inspect and carry, so they work safely next to your people.
- Autonomous floor-cleaning machines on a schedule
- Inspection robots and drones for roofs, tanks and yards
- Autonomous mobile robots moving pallets in a warehouse
- Fleet dashboards with alerts when a unit stalls or goes offline
How we build it
We select proven robot platforms rather than build hardware, connect them to your building network, maps and work-order systems, and monitor the fleet from one place.
Guardrail Robots join a separate network with least-privilege access, run inside defined safety zones, and stop on the manufacturer's safety systems, which no software of ours overrides.
Agents that answer calls and messages at any hour and hand off to your staff when a person is needed.
- Reception: answer, route and take messages
- Booking and rescheduling appointments
- After-hours triage that separates urgent from routine
- Website chat that answers from your own service information
How we build it
The agent connects to your phone system, calendar and customer records, runs on a private or hosted model chosen for the data involved, and transfers to a person with the conversation summary.
Guardrail Callers are told they are speaking with an AI, recordings follow state consent rules, and the agent cannot take payment or change records without a person approving.
Agents that do the repetitive paperwork of finance, operations and administration.
- Reading invoices and matching them to purchase orders
- Sorting and answering shared mailboxes
- Preparing routine reports from several systems
- Routing approvals to the right person with the facts attached
How we build it
Each agent is built for one task, integrated with your accounting, email and document systems, and run in your environment or a private one we operate, with a run-alongside period before it works alone.
Guardrail Every agent has its own narrow identity and a full action log, and anything that moves money, sends to customers or deletes data waits for human sign-off.
Models that read machine data and flag wear before it becomes a breakdown.
- Vibration and temperature trends on motors, pumps and fans
- Energy-use anomalies that point to failing equipment
- Refrigeration and HVAC alerts before a unit fails
- Maintenance tickets opened with the supporting readings
How we build it
We collect readings from existing controllers or new low-cost sensors, run the models at an on-site gateway or in your private cloud, and feed alerts into your maintenance system.
Guardrail Sensors and gateways sit on a segmented operations network with no inbound internet access, and the system advises; it never commands equipment.
Cameras and models that check every item on a line instead of a sample.
- Surface defects, scratches and cracks
- Missing, misplaced or wrong parts in assemblies
- Label, print and packaging checks
- Counting and sorting by size or color
How we build it
We fit cameras and controlled lighting at the line, train the model on your own good and bad examples, and run it on an edge computer wired to a reject signal or an operator screen.
Guardrail Images stay on site, rejected items are kept for review, and we report measured error rates instead of promises.
Anonymous counts that show how people use a store, lobby or venue.
- Footfall and peak-hour patterns
- Queue length with staffing alerts
- Shelf gaps and out-of-stock detection
- Dwell time in areas such as entrances and displays
How we build it
Models run on existing or added cameras at the site, turn video into counts and heat maps on the spot, and send only those numbers to your reporting tools.
Guardrail We design for anonymous counting without identifying individuals, post signage where required, and avoid face recognition.
Scheduling and routing that fits the right technician and parts to each job.
- Job-to-technician matching by skill and location
- Route planning with live changes
- Parts and van-stock suggestions
- Automatic customer updates and arrival windows
How we build it
We connect to your work-order, GPS and calendar systems, build the optimizer around your real rules, and let dispatchers review and override each plan.
Guardrail Location data is limited to work hours and the people who need it, and dispatchers keep the final say.
An assistant that answers staff questions from your own documents without sending them to a public service.
- Policies, procedures and contract questions with source links
- Technical manuals for field and support staff
- Past proposals and project notes found in seconds
- Onboarding answers for new employees
How we build it
We index your documents on a private or on-premises model, respect the permissions each person already has, and show the source passage behind every answer.
Guardrail Documents and prompts stay under your control, access follows your existing permissions, and answers cite sources so people can check them.
