From opportunity to the factory floor.

One shared operational record

When production, inventory, attendance and pay preparation live in separate tools, the thing connecting them is usually a person. Tickets get printed, glued to sheets and scanned the next day. Weekly efficiency is worked out by hand, looked up in a pay-rate table and keyed in again. Punches are pulled off each clock one at a time. The office sees what happened after the floor has already moved on.

This project brings those workflows into one shared manufacturing system on a server in the building. Production records connect to materials, purchasing, quality and traceability. Bilingual browser screens support the people using it on the floor — Spanish by default, because that is the language the floor works in, not a translation layer bolted on later — while management works from the same underlying records. Payroll preparation covers time and piece-rate calculation, and exports into Sage 50, which remains the accounting ledger.

Payroll ran in parallel against the hand calculation until the numbers matched. Only then did it go live.

Finding the next opportunity

The ERP can search for solicitations automatically through the SAM.gov API and configured opportunity-data providers. Scheduled checks bring matching notices into a shared review queue, keep source links and deadlines with the record, and alert the team to new leads. Reviewers can follow changes without rebuilding the search each morning, while the decision to pursue an opportunity and approve a bid stays with people.

A tablet at the machine

Machine operator tablets bring the system to the point where work happens. An operator signs in, sees assigned work, scans a bundle or operation code and records production through a focused Spanish-first screen. The station is linked to its machine, and supported work entries can wait in a local queue when the connection drops, then return to the server for validation.

When something stops the work, the operator can call for a mechanic, quality inspector or materials from the same tablet. The request carries the station context, shows who has acknowledged it and returns a response to the operator. Mechanical calls connect to maintenance work orders and downtime; quality calls connect to the inspector's record. That gives the floor and the office a shared account of what happened and who followed through.

Seeing the floor as it runs

Floor displays keep production KPIs visible during the shift: recorded output, efficiency, work waiting between operations and quality holds. The boards refresh automatically, so supervisors and operators can see where attention is needed as work is recorded. The same underlying records support the supervisor's working screen and the read-only display on the wall.

Machine tracking adds a view of the equipment itself. Sensor controllers running custom firmware send readings from equipped machines into the ERP, including stitch counts, machine activity and cycle timing where the installed sensor supports them. The work includes infrared and vibration-sensing approaches, with calibration for the machine and buffered readings that can be delivered after a connection returns. This connects the physical equipment to its operating history and helps the team investigate interruptions and changes in performance.

Knowing who is in the building

The visitor kiosk turns reception's sign-in book into a usable record: arrival and departure times, the person being visited, photographs, a printed badge and any escort requirement. Hosts are notified, and the office can produce a visitor log for a chosen date range. Missed check-outs are flagged for follow-up. These features support the visitor escort and physical-access logging controls in NIST SP 800-171 as part of the company's wider security program.

The employee timekeeping tools also include a local facial-recognition option, with documented consent, protected enrollment records and a PIN step when faces are too similar to distinguish. A keypad route keeps time recording available when a match cannot be made. Visitor identity checks remain a staff responsibility.

AI on the company's own hardware

The AI layer is being built around local GPU processing, so selected work can run on hardware the company controls. A local assistant is built to answer staff questions from the ERP's manuals, with references they can check. Custom AI tools for document analysis and search are being connected to this work. They are designed to run on local GPUs, read supplier paperwork, find relevant passages and prepare answers linked to their sources. The purpose is to put useful information beside the task someone is trying to finish.

The document intake also accepts email items and ZIP packages. It sorts and tags incoming paperwork by contract and document type, so the team can find what belongs to a job instead of rebuilding the trail across inboxes and folders. The documents become part of the working record, ready for the next person who needs them.

The approach combines local models with external assistance for selected, approved tasks. What can be shared is defined for each task, rather than giving a model unrestricted access to the ERP. Controlled technical documents stay in their separate protected environment.

From CAD files to draft patterns

The work around the ERP extends into engineering. Custom CAD and pattern-analysis tools read supported CAD exports and pattern files, prepare draft pieces from drawing geometry, check dimensions and seam allowances, and propose size variations for a pattern maker to review. That creates a starting point for skilled work, with discrepancies visible before a piece is approved for cutting. These tools are being developed alongside the ERP; the full pattern workflow is still being connected.

Understanding what drives the cost

The ERP's costing starts with the bill of materials, supplier prices, material usage, freight and allowances. The next step is AI analysis built on those calculated figures: explaining why a quote changed, comparing material or usage assumptions, and drawing attention to missing prices before someone commits to a bid. The calculations remain in the business system, and the estimator remains responsible for the decision.

From the invoice inbox to a traceable accounting record

The accounts-payable automation began in the cloud. It reads invoices from incoming email, captures the source documents, and prepares a review workbook with the supplier's existing vendor ID, proposed general-ledger coding and purchase-order reference where available. Accounting resolves missing information and approves the invoices before a Sage 50 import file is produced.

The next step is to bring that AP agent into the ERP when the local GPU is installed. The ERP's vendor records already connect accounting IDs and GL defaults to purchasing. Bringing the invoice process into that same system is intended to preserve the full trail: the original email and document, vendor and PO, receipts and material use, review decisions, and accounting export. Unclear links need review, and a captured invoice still needs accounting approval before it can be posted.

The project shows what it takes to build around an established operation: preserve the rules people depend on, connect the records, and make the next task easier to finish than it was before.