The questions manufacturers ask before a demo

Two weeks from the day you send work instructions, on a station with one operator. Where several people work the same bench, plan for closer to three. Assembler AI supplies and provisions the camera kit, and nothing about the line has to change.

Day one you send work instructions and Operator Vision reads them. Week one the camera goes up. Week two is calibration and training the model on your alerts, where we compare the footage to the document and settle the differences with your team. Alerts go live on day fifteen.

That’s common on at least a few stations in most plants. Assembler AI will work with you to create them as part of getting that station live.

We assume they might be. Calibration is where that surfaces, and what usually turns up are ghost steps, written instructions no one has performed in years, alongside steps your operators do that the document never captured. Most customers end up with better work instructions before a single alert goes out.

One. Not because we think small, but because you shouldn’t instrument a plant on a vendor’s promise. Two weeks later you have data on your own parts and you can decide what happens next. Stations after the first go up in parallel, so covering a line or a plant is measured in weeks rather than one station at a time.

No. The camera mounts above the station, nothing gets bolted to your tools, and nothing about the bench changes.

If a person standing at the station could see it, so can Operator Vision. Steps, sequence, parts, tools, placement, counts, dwell time, and timing, on every cycle rather than on the ones somebody sampled.

What sits inside a closed housing, which isn’t visible to a supervisor standing there either. Very small parts and fine markings depend on camera position, and Assembler AI confirms what’s resolvable during calibration rather than after go-live. Values that live in a tool, like a torque reading, come from the tool.

One camera covers a working area of roughly ten feet by ten feet. Larger stations are handled with multiple cameras, and we’ve deployed that way on assemblies well beyond fifty feet. Where an operator works both sides of a unit, an ATV or a truck bed, that station also gets more than one. Assembler AI works out the coverage with you at the site visit rather than leaving you to find a gap in week three.

That’s the environment Operator Vision was built for. Where your variants look visually different, it identifies which one is in production and switches automatically as the station changes over, across as many as ten configurations on the same line. Variants that look nearly identical get taught during calibration.

Assembler AI’s proprietary manufacturing vision models start around 80 percent accuracy on a new station and reach the mid 90s after the initial calibration. Those first two weeks are what get it there on your parts rather than in a demo.

As many as are worth acting on, which is usually fewer than people expect. Operator Vision only sends what it has high confidence in, and Assembler AI works with your team to set which events matter and who receives them. If something important isn’t getting through, we open the threshold. If a rule is firing on noise, we tighten it.

Operator Vision handles most changes on its own, including lighting shifts and minor differences in parts. New work instructions or a complete layout change need additional training, which Assembler AI takes care of for you. New variants it has to tell apart, a left-hand against a right-hand version, get taught the same way.

No. Faces are blurred, we don’t collect operator names or IDs, and the analysis sits on the process. Events come back tied to a station and a shift, not to a person.

The operator, first. The alert reaches the station while the piece is still there and the correction costs a few seconds. Most alerts end there, which is the entire point. If it isn’t handled at the station it escalates to a supervisor with the video attached, and either way the event lands on the dashboard.

Say so from the station screen. That feedback goes straight into training Operator Vision, so the model gets more accurate on your station every time an operator weighs in.

No. Variance shows up as variance between runs and between shifts, tied to a station. That’s the version that leads to a better standard rather than a difficult conversation.

Operator Vision runs in unionized plants. Faces are blurred, no operator names or IDs are collected, and nothing tracks an individual operator. Agreements vary, so Assembler AI works through the conversation with your union representatives alongside you, and the deployments that go smoothly start with everyone in the room before the camera goes up.

That depends almost entirely on how it’s introduced, which is why we would rather be in the room early. The feedback reaches the operator before anyone else, and the station screen gives the bench a route to engineering it usually doesn’t have.

No, and that’s the difference from a connected worker platform. Those depend on operator inputs, a tap or a click or a keystroke, which means stopping the build to record it. The camera collects the same record without the operator doing anything.

One outbound path. The camera sits on its own subnet and streams out over an encrypted VPN. Nothing connects to your production network.

Video streams over RTSP inside an encrypted VPN tunnel, and everything at rest is AES-256 under a customer-managed key issued per tenant. Data is logically isolated per customer.

You own it. Production data sits in a US or Canadian cloud region and you choose which at deployment. Default retention is twelve months, and you can set your own, delete earlier, or export at any time.

Off-the-shelf, NDAA-compliant cameras that Assembler AI supplies and provisions, with no restricted foreign-made surveillance equipment in the stack.

Yes, and it’s written for IT and legal rather than for marketing. Request the security overview without booking a call.

Raise it in the first conversation. NDAA-compliant hardware, US data residency, and a single outbound path cover most of what a supply chain review asks for. If your programs carry specific certification or handling requirements, tell us and we’ll be straight with you about where we stand against them.

Role-based permissions determine who can view footage and dashboards. If you’re running more than one site, tell Assembler AI early and we’ll scope single sign-on and per-site data separation with your IT and HR teams while the first station runs.

No. Operator Vision runs standalone and your work instructions are the only required input. Assembler AI provides MES and ERP integrations when you want data flowing both ways.

Yes, the data as CSV and the video alongside it. Dashboards cover station, shift, step, and SKU, with side-by-side comparison of two time periods.

Yes, and most customers start with one. All four run on the same camera and the same feed, so adding another later is a change in what you’re subscribed to rather than another deployment.

That works and Assembler AI scopes deployments that way. Quality and risk alerts run independently of the productivity reporting, and you can leave productivity switched off.

One per station where you want operator guidance. Stations running quality alerts or cycle time analytics don’t need one, since those report to supervisors and dashboards.

Step-level data on that station, every cycle and every shift, with video on every outlier and every alert. Enough to say which steps get missed, how often, and on which shift.

Coverage and accuracy. A manual study covers a handful of cycles on a handful of stations, and the operator works differently while it’s running. Operator Vision covers every cycle on every shift, and after about a week the camera stops being new and the data starts describing the job rather than the audit.

Yes. Log the change itself, extra training for an operator, a new tool, a reordered sequence, and Operator Vision reports what happened to cycle time afterward, measured against real production data rather than another study.

It gets shorter. Operator Vision holds timestamped video for every flagged cycle, indexed by step, so an analysis that used to hold a room until eight at night becomes a lookup.

Because the data is comparative. One station tells you how that station runs. Ten tell you which runs best and why. Across shifts, cells, and sites the same work becomes directly comparable, and a group of plants gives you a benchmark that keeps updating itself as your work changes.