EC
All projects

04 · Python · FastAPI · React

Heating Ring Design Automation

A customer's layout-diagram JSON goes in, and a full production package comes out: bill of materials, winding programs and sheet-metal DXF.

Gienkee Plastics Machinery, Engineering Intern (part-time), May 2026 to present

  • Python
  • FastAPI
  • React
  • Vite
  • TypeScript
  • Gemini API
Heating Ring Design Automation wizard showing calculated segments
Live captureStep 2 of the three-step wizard on the real app: segments calculated from an uploaded layout spec, editable before generating the production package.

What it replaced

Gienkee Plastics Machinery makes heating rings, and every order used to start from a hand-filled spec sheet. This is the internal web tool I built to replace that. A layout-diagram JSON ({machine, specs}) goes in, and everything downstream is derived through deterministic calculation engines: bill of materials, winding parameters and CNC winding programs, and sheet-metal DXF, merged into a single production-package PDF plus DXF.

It runs as a three-step wizard (upload spec, review and edit the calculated segments, generate and download the package) or a CLI, and the core workflow runs end to end in production today.

Reading the spec off the customer's PDF

Customers hand over layout diagrams as PDFs, often scans with no text layer and sometimes password-protected. An ingestion module feeds those PDFs to Gemini for structured extraction, producing the same {machine, specs} JSON a hand-filled spec would, so the result goes through the identical downstream validation and rendering path with no separate parsing code to maintain.

Every extracted JSON is still checked against the same deterministic schema validator. Where Gemini is not confident about a field, it still fills in a best guess but records a {"field": ..., "reason": ...} entry in a top-level _review array. Validation and rendering never read that array. It exists so a person can scan it before trusting the extraction.

By the numbers

On the factory floor the blocker turned out to be unmatched cycle times between concurrent steps, which left fragmented idle time. Speeding that up cut cycle time 20% on average across all variations.

20%
faster cycle time on average, all variations
7.2K
lines of Python, plus a React and TypeScript front end
204
tests: 186 Python, 18 TypeScript
2
entry points: three-step web wizard and CLI

Deliberately not built yet

The project's own docs are explicit about scope. There is no automated .docx intake for the customer information sheet, no independent rules-engine module (the BOM, winding and sheet-metal derivations are deterministic formulas living in their calculation modules), no SolidWorks two-way sync, and no single-file distribution that skips a Python and Node install. The calculation engine has been reverse-derived from one machine series so far, and other series do not have real parameters yet.

Design principles

The calculation engine is pure deterministic formulas with no model inference, fully overridable and fully offline. PDF extraction is the one place an LLM touches the pipeline, and its output is never trusted directly. It is validated the same way a human-entered spec would be, with uncertainty surfaced rather than hidden.

If this is the kind of work you need, let's talk.

Coffee chat