1. A photo of a broken part
One picture of the snapped piece is enough. SynaCAD reconstructs the geometry, proposes a stronger version, and exports a 3D-print-ready file (STL or 3MF) you can send to any printer.
Currently in development · Open-source · MIT licence
SynaCAD turns a photo, a sketch, or a few sentences into a 3D-print-ready file or a full manufacturing drawing pack.
Every number is computed by a validated solver, never invented by the language model. in development
What SynaCAD does
Tell SynaCAD what you need in whatever form is easiest. Get back something you can actually print, machine, or hand to a manufacturer.
One picture of the snapped piece is enough. SynaCAD reconstructs the geometry, proposes a stronger version, and exports a 3D-print-ready file (STL or 3MF) you can send to any printer.
"A 6 mm 6061-aluminium L-bracket, 80 mm tall, two M5 holes 50 mm apart, must hold 30 kg." SynaCAD sizes it, picks the wall thicknesses, and outputs the CAD plus a fully-dimensioned drawing.
Draw the rough shape, photograph it. SynaCAD interprets the lines and turns them into a parametric CAD model you can refine in conversation, then export.
STEP or STL in. SynaCAD audits the design, finds lighter / cheaper / more printable variants on a Pareto front, and emits the analysis paperwork to back the change.
Imagine a child snaps a photo of a broken plastic clip on a chair. Within minutes, a 3D printer produces a stronger, lighter replacement that the chair will outlive. SynaCAD scales from that clip up to a steel bracket holding industrial loads, or a wind-turbine mount — same workflow, same honesty about the numbers.
Who it's for
Same agent, same physics, same standards. The output adjusts to who is asking.
Broken IKEA bracket. Lost knob on a 12-year-old appliance. A clip that snapped on the school chair. Photograph it; print the replacement; move on.
An MVP enclosure designed and printable in an afternoon. A custom mount sized for the load it actually carries, with a sketch-to-shop drawing you can email a contract manufacturer.
Material-by-material design audits of existing parts. Full GD&T drawing packs traceable to ASME Y14.5. Inverse design across mass, cost, and life — with citations every reviewer can check.
How it stays honest (design principles, in active development)
The language model is a router, not an oracle. It does not make up numbers.
Every dimension, tolerance, and material call comes from a validated solver — the same kind of code a senior engineer would have written by hand.
The engineering report cites the standards it relied on — Bruhn, Niu, ESDU, Lekhnitskii, ASME Y14.5 — so a reviewer can check the working without having to trust the model.
Before a file leaves the agent, it passes a sanity check — strength, fatigue, printability, tolerance feasibility. Anything that fails is flagged, not hidden.
If SynaCAD does not have a validated way to answer your question, it says so. It does not guess.
What you get back
An STL or 3MF file ready for any 3D printer, plus a short note on how strong the part will be and what it can take.
An ASME Y14.5-compliant, GD&T-annotated drawing pack — the document a contract manufacturer asks for.
An analysis report citing every relevant Bruhn / Niu / ESDU equation. Plus the CNC G-code or printer slicer file, ready to send.
12-month roadmap
STEP/STL parsing, photo-to-CAD reconstruction, conversational CAD generation, feature extraction. Open-sourcing the underlying classical sizing tools.
LLM dispatcher, with constraint enforcement happening before any output leaves the agent.
Pareto-front optimisation across mass, cost, and life. ASME Y14.5 GD&T generation.
Public v1.0 with four worked examples — a kitchen-chair clip rebuilt from a photo, a stress bracket, a fatigue lug, and a riveted lap-joint coupon.
Track record
SynaCAD is a new project, not a remix of these. But these are some of the things I have already built, validated, and shipped, and combined with hands-on experience training deep-learning models, they are the reason I think this work will land.
A Lekhnitskii anisotropic stress-around-hole solver I wrote during my MTech and shipped to a company through Aeroknacks. Now MIT-licensed and public.
github.com/allamaprabhuani/bjsfm →A GPU-native, end-to-end-differentiable phase-field fracture solver in PyTorch built during my PhD. Roughly 8× faster than the standard reference codes on a single GPU; validated against four of them (Akantu, FEniCSx, COMSOL, PhaseFieldX).
Lug strength, Cozzone plastic-bending, fastener load-transfer, ABD matrices, column buckling. Built during my MTech, used in real engineering practice, rolling into open source as SynaCAD absorbs them.
Hands-on experience training ML models for fatigue-life prediction in aluminium alloys, plus first-author on a review of ML for computational fracture mechanics with collaborators at EPFL and Florida.
Why "Synapse"
A long-form post on what SynaCAD is, what it isn't, and the design decisions behind it. The synapse is where a signal becomes a decision — between the classical mechanics canon (Bruhn, Niu, ESDU, Lekhnitskii) and modern generative models that can read your sketch, your photo, your sentence.