Home

Welcome to the Sofia Engine Wiki

Sofia Engine is an engineering-grade, offline-first General Scientific AI & Edge Intelligence Runtime developed by Rootcastle Engineering & Innovation for physical telemetry, condition monitoring, multi-domain digital signal processing (DSP), quantum computing emulation, low-level assembly neural self-training, and safe technical automation.

This wiki provides comprehensive mathematical formulations, architectural blueprints, deployment guides, and API contracts for engineers, researchers, and systems architects deploying Sofia Engine on edge gateways, bare metal, or microcontrollers.

---

The Rootcastle Engineering Pillars

Sofia Engine was engineered from the ground up to replace fragile, stochastic cloud prototypes with deterministic edge infrastructure:

+---------------------------------------------------------------------------------------+
|                                ROOTCASTLE PILLARS                                     |
+---------------------------------------------------------------------------------------+
|  1. GENERAL SCIENTIFIC AI      Unifies physics, DSP, quantum emulation, and ML into   |
|                                a rigorous, reproducible scientific framework.         |
|  2. ASSEMBLY-LEVEL AUTONOMY    Self-training neural models running on virtual/native   |
|                                assembly with zero framework overhead.                 |
|  3. AUTOMATED FINE-TUNING      Curates telemetry into JSONL datasets and triggers      |
|                                fine-tuning via NVIDIA NIM and OpenRouter APIs.        |
|  4. EVIDENCE BEATS HYPE        Every metric derives from verifiable physical/spectral |
|                                evidence. No unbacked accuracy claims.                 |
|  5. DEFAULT "DENY" SAFETY      Zero control path bypass. All control decisions pass   |
|                                through physical interlocks and operator gates.        |
|  6. BOUNDED RESOURCE ENVELOPE  Strict ceilings on buffers, queues, windows, and disk. |
|                                Zero unbounded memory allocations.                     |
|  7. AIR-GAPPED BY DESIGN       Zero network or broker dependency in the core. Runs on |
|                                bare metal, isolated gateways, and microcontrollers.   |
+---------------------------------------------------------------------------------------+

---

Wiki Navigation & Documentation Map

Explore the detailed engineering manuals:

Section	Document	Focus Areas
Core Architecture	System-Architecture	Ingest buffers, signal pipeline, inference registry, diagnostic engine, edge persistence.
AI Copilot & Providers	AI-Engine-&-Copilot	NVIDIA NIM, OpenRouter, and deterministic offline decision support.
Assembly Neural Engine	Assembly-Neural-Engine	In-situ assembly VM, self-training neural model, SGD backpropagation, x86/ARM/WASM emitters.
Auto Fine-Tuning	Automatic-Fine-Tuning	Telemetry dataset curation, JSONL generation, automated API fine-tuning, checkpoint registry.
Multi-Domain Signals	Multi-Domain-Signal-Processing	Electrical power quality (IEEE 519), acoustic emission (ASTM E1316), fluid, and IMU kinematics.
DSP Foundations	Digital-Signal-Processing	Welch PSD (Parseval conservation), Hilbert envelope, rotating machinery kinematics (BPFO/BPFI).
Quantum Emulation	Quantum-Computing-Emulation	Complex statevectors in $\mathbb{C}^{2^n}$, universal gates, quantum kernel estimation.
Machine Health	Condition-Monitoring-&-Machine-Health	ISO 10816/20816 vibration severity, robust MAD/EWMA/CUSUM anomaly detection, health scoring.
Control Safety	Safety-Gate-&-Policy-Engine	Default DENY state machine, operator approval, interlocks, Nonce/TTL replay defense.
Embedded & TinyML	Embedded-C99-&-TinyML	C99 reference runtime, zero heap allocation post-init, Q16.16 fixed-point math, golden-vector tests.
Field Protocols	Telemetry-Adapters-&-Protocols	CSV, JSONL, In-Memory, Replay, MQTT (TLS/QoS), Modbus TCP/RTU, RS-485 Serial adapters.
Operations	CLI-&-Developer-Operations	CLI commands (sofia doctor, analyze, benchmark), testing suite, typing & linting workflows.
Security	Security-Policy-&-Threat-Model	Threat model covering 21 edge threats, safe deserialization (no pickle), Ed25519 model signing.

---

Quick Architecture Summary

flowchart LR
    A["Multi-Domain Telemetry\n(Sensors/Vib/Elec/Acoustic)"] --> B["Bounded Ingest\n(Ring Buffer / Drop-Oldest)"]
    B --> C["Multi-Domain DSP\n(Welch / Hilbert / IEEE 519)"]
    C --> D["Assembly Self-Training & Inference\n(SofiaAsmVM / Quantum / MAD)"]
    D --> E["Diagnostic Engine\n(Evidence Fusion / Quality Scaling)"]
    E --> F["Policy Gate\n(Default: DENY)"]
    F -->|Approved| G["Actuator / PLC"]
    F -->|Denied| H["Audit Log / Safe Hold"]
    E --> I["AI Copilot & Auto Fine-Tuning\n(NVIDIA / OpenRouter / JSONL)"]

Loading

---

Getting Help & Contributing

• Codebase & Issues: GitHub Repository
• Company & Research: Rootcastle Engineering & Innovation
• Security Advisories: Report privately per SECURITY.md