Project Deep Dive
My Assembly Hub
A local-first manufacturing intelligence system built to close the gap between machine exports and engineer decisions on a live SMT line. One intake gate. One SQLite database. One browser UI. 300+ production jobs and roughly half the programming time.
Job Loader
The entry point. Every downstream module lives or dies by what comes in here.
Process Flow
The Problem
Before Job Loader existed, engineers navigated a tangle of machine folders and vendor exports on multiple workstations — no single canonical source of what a job actually contained. The wrong revision of a file would silently propagate to AOI analysis, rotation logic, and test validation. The damage only surfaced later, on the floor.
The Approach
Job Loader is the one door into the database. It parses the Excel BOM and, optionally, the CAD placement file, then validates every row: duplicate designators, BOM-versus-placement conflicts, part numbers the library does not know. A validation token holds the parsed rows while the engineer resolves each problem row, and fixes apply as a batch that is rejected whole if any single action is bad. Only then does the BOM, with its placement rows, commit in one transaction. Placement, AOI, FPT and the dashboards all read that clean record rather than raw vendor files.
Why This Solution
Making Job Loader the mandatory gate rather than an optional convenience was the key architectural decision. A problem row that reaches the database silently becomes a wrong rotation, a wrong AOI line or a false FPT failure three steps later, where it costs a board instead of a click. Holding the rows behind a token until a human has resolved every problem, then committing atomically, means downstream modules never see a half-imported job.