An on-premises platform that joins the AOI images and calls, unit and lot identifiers, and equipment events your line already produces into one searchable quality record. It enters through the problem you feel first, AOI false-call review, and never sends a board image off-premises.
The AOI holds the images and its own calls. The operator who overrules a false call records it, if at all, somewhere else, and the AOI forgets the correction the moment it is made. Serial, lot, and work order live in the MES or a spreadsheet. Placement, reflow, and test keep their own logs in vendor formats. When a defect is confirmed, in the plant or in the field, finding out which other units were exposed, and what inspection saw at the time, is manual record assembly. And in high-mix production the false-call burden that feeds this problem never settles: every new product family restarts the tuning, so review labor compounds with product diversity rather than diminishing with experience. None of this is recoverable after the fact — a plant that isn't retaining its images, verdicts, and overrides today has no history to analyze tomorrow, and no AI or ML tool, ours or any vendor's, can be trained or evaluated against a history that was never kept.
Every module reads and writes one record: unit, event, genealogy, evidence. Connect brings the data in, Review clears the false calls, Curate turns your decisions into a dataset and model you own, and Trace answers the question a confirmed defect raises. No AOI hardware is replaced, no production process is modified, and no data leaves the building. The record itself is the precondition, not a byproduct: it starts accumulating the day Connect goes live, and it's what makes every later improvement — false-call trend analysis, drift detection, your own retrained model, or any tool you adopt after us — measurable instead of guesswork.
Ingests the flagged-board images and results your AOI already exports, plus placement, reflow, and test events over IPC-CFX, OPC-UA, MQTT, or file exchange. Every record is keyed to unit, lot, work order, station, and model version. Read-only against every source system.
AOI export · IPC-2591 CFX · OPC-UA · MQTT · fileAn on-appliance classifier clears high-confidence cases in under 100 milliseconds against thresholds you own. Uncertain cases escalate to a locally hosted vision-language model that reasons against IPC-A-610 in plain language. What remains goes to your operators' review queue. Every verdict is written against the unit.
Classifier · local VLM reviewer · human queue · the entry productEvery operator decision is retained as a labeled example. Your quality engineers curate, retrain, approve, and roll back on an on-site workstation you own. The dataset and the model are yours, and they never leave the facility.
Versioned dataset · model registry · one-click rollbackUnit and lot genealogy, defect trace-back to affected units and process windows, and an exportable evidence package aligned to IPC-1782 traceability levels, for a prime contractor's corrective-action request or a medical device customer's nonconformance investigation.
Containment · unit history · evidence exportADVISORY ONLY — THE PLATFORM NEVER ENTERS THE PLC REAL-TIME CONTROL LOOP.
Nobody is asked to trust a black box. Thresholds are yours to set, evidence is yours to inspect, model versions are yours to approve or roll back, and the unit record is yours to query and export.
Both verticals share the same structural condition: regulation keeps cloud AI off the production floor, and the overlay products on the market are built for the high-volume, cloud-connected shops on the other side of that line. One platform serves both — and market entry is sequenced by design, not by accident.
Validated processes, documented change control, and complete traceability are contractual requirements in medical device production — not preferences. The platform's decision audit trail, model version history, escape ledger, and one-click rollback map directly onto that validation culture, with no product modification.
The air-gap-native architecture and clean software bill of materials are designed for precisely this environment. ITAR permits cloud processing of controlled technical data only inside U.S.-persons-only environments or under equivalent access controls — out of reach for general commercial cloud infrastructure, which is why the AI overlay products this segment can actually buy are the ones built to run entirely inside the building.
This platform cannot be delivered by a machine learning specialist, a data engineer, or a controls engineer working in isolation. It demands simultaneous, production-grade fluency across every discipline it touches — assembled here in one person, with zero communication loss between disciplines. The 70–90% false-positive reductions reported for AI overlays in cloud-connected plants are an industry benchmark, not our result; our results will be the ones your pilot measures.
For quality engineering leaders at high-mix, compliance-constrained EMS providers — defense and aerospace under ITAR, and medical device under ISO 13485 and FDA design controls — who cannot readily adopt general commercial cloud AI and cannot afford an enterprise traceability system, the platform adds an on-premises, air-gapped data layer on top of the AOI and production equipment you already own — built to published industry standards rather than one vendor's export — compliant by architecture, transparent in every decision, owned by you at contract end, and supported in person, in your region, the same day.
The US Southwest is becoming the nation's semiconductor and electronics manufacturing center: TSMC's Phoenix campus is operational and expanding, Intel is investing $20 billion in Chandler, Amkor is constructing an advanced packaging facility in Peoria announced at $2 billion and since expanded to $7 billion, and more than 35 ecosystem companies have announced Arizona expansions — over $100 billion in combined regional investment. That build-out feeds the regulated EMS supply chain this platform serves, at exactly the moment those shops are competing for scarce skilled labor and facing rising quality-scorecard pressure. The same obligations and the same gap exist in every established U.S. defense, aerospace, and medical device electronics cluster — Southern California, Texas, the Colorado Front Range, Huntsville, the Florida Space Coast, Minneapolis, New England. Phoenix is where we start; the platform is built to published industry standards so the next plant, in the next region, adopts the same data model without a rebuild.
Leave your details and we'll keep you informed as the pilot program opens in your vertical — measured results, published pricing, no sales pressure. Here is how a pilot works, in writing, before anyone asks for your time:
Occasional updates on pilot availability and measured results. No spam, and your details are never shared.