Rigel

01The question

Aerospace design workflows suffer from severe friction between high-level engineering intent and low-level computational geometry synthesis.

Engineers waste days manually modeling CAD brackets, configuring meshes, and guessing parameters, while generative AI hallucinations cannot be trusted in safety-critical airframes.

02Why it mattered

Every aerospace structural component must satisfy rigorous bending, shear, and stress-concentration limits under single or multi-axial load cases (e.g. Al6061-T6 under 4 kN).

'AI handles ambiguity. Math handles precision. These roles never swap.'

03How we approached it

The Derivation Engine: First-principles physics derivation where local LLaMA 3.2 3B extracts constraints, Shigley equations derive dimensions without hardcoding, PicoGK synthesizes B-Rep geometry, Gmsh meshes with Tet10 volume elements, and CalculiX static FEA verifies Von Mises convergence.

04How it works

Constraint-driven aerodynamic intelligence engine synthesizing production geometry directly from load specifications with cryptographic SHA-256 audit chaining.

05How it's built

Rust control plane with C/C++ FFI bindings (Gmsh C API, CalculiX subprocess).

Outputs four mandatory production artifacts every run: bracket.step, bracket.dxf, report.pdf, and audit.json.

• Rust
• Shigley Physics
• PicoGK Kernel
• Gmsh (Tet10)
• CalculiX FEA
• SHA-256 Audit Chain
• Al6061-T6 Sizing

06What we observed

On an 8-core workstation, the derivation pipeline executes in 8-12 minutes without CAD hallucination.

Physical bracket testing scheduled on workshop press at PSG College of Engineering, Coimbatore, benchmarking predicted vs actual failure loads (target < 15% error, logged in public FAILURES.md).

07Where it is now

Stage 1 MVP in active development.

Six-file architecture discipline strictly enforced; core solver lattice running local-only computation with verified Shigley equations.