JARVIS.From intent toengineering execution.
Not a chatbot. An engineering AI and control system that turns what an engineer means into what a system does.
01 Engineering Execution Graph
Eight stages from intent to action
Plain language, sketches or requirements.
Objects, relationships, quantities, purpose.
Domain knowledge and trade-offs.
Loads, tolerances, materials, standards.
Editable parameters and relations.
Parameters drive CAD operations.
Checked against constraints.
Files, operations, fabrication steps.
02 In practice
What a request looks like
An engineer states a requirement. JARVIS parses it, applies domain reasoning, makes the constraints explicit, produces an editable parametric model, drives CAD to create geometry, and validates the result before anything is accepted.
Every stage is inspectable, so a mistake can be corrected at the level where it was made.
# illustrative — not a recorded session intent "Mounting bracket, 120 mm, two M6 holes, aluminium" semantic part=bracket length=120 holes=2×M6 material=aluminium constraints edge_distance ≥ 1.5·d tolerance ±0.1 parametric length, hole_pitch, thickness → editable geometry CAD operations → STEP validate constraints checked ✓
03 Architecture
Four cooperating layers
Parser layer
Turns requests and documents into structured engineering objects.
Planner layer
Decides which engineering steps are needed and in what order.
Engineering layer
Applies calculations, constraints, repair and provenance tracking.
CAD layer
Produces real geometry and exports engineering files.
04 Positioning
A chatbot versus an engineering system
- Produces text from a prompt
- Constraints stay implicit
- Output is hard to verify
- Typically cloud-dependent
- Produces parameters, geometry and files
- Constraints are explicit and checked
- Deterministic validation before acceptance
- Designed local-first
→ Contact Zaftrus
Let's build something that matters.
Research collaborations, technology conversations, AEGIS NEXA and JARVIS inquiries, ecosystem and media — every message is read by the Zaftrus team.