Show HN: A neuro-symbolic manufacturing engine built in 1 week with Gemini 3.0 (github.com)

OpenForge is a one-week research prototype demonstrating a neuro-symbolic manufacturing engine that turns natural-language intent into verified hardware artifacts — BOMs, blueprints and browser flight simulations — by combining Gemini 3.0 LLMs with deterministic physics and supply‑chain logic. The system uses a multi-agent pipeline to source parts from the web, a “refinery” agent (Playwright-driven) to scrape and complete missing specs, vision models to read PCB silkscreens and drawings, and an AI assembler that procedurally generates and catalogs viable drone assemblies. It’s notable because it moves beyond text-only LLM workflows to integrate real-world constraints (mass, motor curves, geometry) and supply-chain data, producing digital twins that can be simulated in Three.js + Cannon.js. Technically, OpenForge enforces a Constraint Chain/CompatibilityService that checks voltage vs motor KV, UART counts, prop size vs frame clearance, etc., before any AI design step, while class filters limit combinatorial explosion by bucketing components. Core scripts (seed_arsenal.py, refine_arsenal.py, design_fleet.py, fly_drone.py) orchestrate seeding, active auditing, generation and simulation; requirements include Python 3.10+, Playwright, Google Gemini and Custom Search keys. Limitations: the physics sim and controls are rough, assemblies sometimes misalign and require multiple seed/refine runs, and cost-optimization/feedback loops are still in progress. As a proof-of-concept it highlights how LLM orchestration + symbolic validation can automate hardware design while exposing practical gaps in data quality, tooling and real-world validation.