Marine engineer building the AI — and now the silicon — that runs in the engine room.
From engine room to ASIC. From J1939 to LLM inference.
3+ years hands-on in commercial marine power plants — medium-speed and high-speed 4-stroke diesel engines, mechanical fuel-injection systems (Bosch P-pump rebuilds), turbocharger overhaul and dynamic balancing, fuel-injector pressure testing, scavenge-space and piston-ring inspection, auxiliary genset servicing, planned-maintenance-system execution. The kind of detail you only learn with grease on your hands.
Hands-on with both legacy mechanically-governed engines (Bosch P-pump + pneumatic / Woodward governors) and modern ECU-controlled repower projects (Caterpillar ADEM, Cummins CM-series, J1939 telemetry).
Every project I build came from a problem I saw on board. I solve it at the layer where it actually needs to be solved — sometimes Python, sometimes a Rust crate over J1939, sometimes SystemVerilog targeting Skywater 130 nm.
| Problem on board | What I built |
|---|---|
| Confined-space inspections cost $50–100 K, risk lives | ARGOS — autonomous inspection robot (edge AI + TRIZ reasoning) |
| Unplanned engine failure costs $50–500 K/day | POSEIDON-DIAG — real-time J1939 / NMEA 2000 diagnostics + AI anomaly detection |
| Time-based PMS wastes 30–50 % of maintenance budget | TRITON-ML — RUL prediction 2–4 weeks ahead of classical alarms |
| Operators ignore alarms past 500+ params | AEGIS-MONITOR — 3D ship dashboard with intelligent prioritization |
| IMO 2030/2050 demands radical engineering R&D | SYNIZ — 50 TRIZ agents debating contradictions to compress R&D cycles |
| Satellite uplink is 64–512 kbps, cloud AI doesn't fit | NautilusQuant — 4× deterministic compression + custom ASIC for shipboard inference |
graph TD
A["AEGIS-MONITOR<br/><i>Operator Dashboard</i><br/>Live telemetry · 3D model · Alerts"] --> B
B["ARGOS<br/><i>Inspection Robot</i><br/>Vision · Navigation · Edge AI"] --> C
B --> D
B --> E
C["SYNIZ<br/><i>TRIZ Engine</i><br/>50 agents reason<br/>about the unknown"]
D["TRITON-ML<br/><i>Predictive Maintenance</i><br/>Fault detection + RUL"]
E["POSEIDON-DIAG<br/><i>Ship Interface</i><br/>J1939 · NMEA 2000 · CAN"]
D --> F["NautilusQuant<br/><i>Edge Compression + Custom ASIC</i><br/>21-opcode ISA · 1.5 KB LUT<br/>RTL · Yosys · OpenLane MPW path"]
style A fill:#0d4a6b,stroke:#1e88a8,color:#e2e8f0
style B fill:#6b2d0d,stroke:#a8571e,color:#e2e8f0
style C fill:#2d6b0d,stroke:#4a8a1e,color:#e2e8f0
style D fill:#4a0d6b,stroke:#7a1ea8,color:#e2e8f0
style E fill:#0d4a6b,stroke:#1e88a8,color:#e2e8f0
style F fill:#6b5a0d,stroke:#a8901e,color:#e2e8f0
The robot sees. The ML predicts. TRIZ reasons about the unknown. NautilusQuant fits inference into the satellite pipe — and now into custom silicon. The human makes the final call.
Read the full engineering vision: from fundamental physics to autonomous systems →
| Project | Status | Problem it solves | Stack |
|---|---|---|---|
| NautilusQuant | 🆕 v0.1.0 · 241 tests · pre-silicon | Satellite uplink 64–512 kbps. Shipboard AI without cloud dependency. Now ships full pre-silicon stack: 21-opcode ISA + RTL + OpenLane MPW config. | Python · PyTorch · Triton · SystemVerilog · Yosys · OpenLane |
| ARGOS | 🔄 active | Hull and tank inspections cost $50–100 K and put humans at risk. Edge AI + TRIZ in confined spaces. | Python · Rust · ROS 2 · ONNX |
| POSEIDON-DIAG | 🔄 active | Unplanned engine failure costs $50–500 K/day. Real-time J1939 / NMEA 2000 diagnostics catch failures early. | Rust · Tauri · React · J1939 |
| TRITON-ML | 🔄 active | Time-based PMS wastes 30–50 % budget. ML predicts true equipment condition 2–4 weeks ahead. | Python · XGBoost · PyTorch · SHAP |
| SYNIZ | 🔄 active | IMO 2030/2050 demands radical engineering. 50 TRIZ agents accelerate R&D cycles. | Python · FastAPI · Neo4j · D3.js |
| AEGIS-MONITOR | 🔄 active | 500+ parameters → alarm fatigue. 3D ship dashboard with intelligent prioritization. | React · TypeScript · Three.js |
| Project | Status | Problem it solves | Stack |
|---|---|---|---|
| arc.computer | 🆕 alpha | Engineering knowledge that works without internet, cloud or subscription. Offline AI engineer turns scrap electronics into working tools via reverse-BOM solving. | Python 3.13 · FastAPI · LLM adapters · offline knowledge base |
| DAEDALUS | 🔄 active | Same diagnostics expertise extended to commercial vehicles — AI-assisted ECU reading, DTC analysis and map editing. | Rust · React · Tauri · J1939 · CAN |
Hardware & Silicon
ML / AI / Robotics
Marine Engine Systems
Marine Automation & Industrial Protocols
Edge & Systems
Web & Visualization
Ship Power Plants ██████████ Marine Diesel Engines
Propulsion Systems █████████░ Overhaul & Diagnostics
Auxiliary Machinery █████████░ Pumps · Compressors · Heat Exchangers
Engine Control ████████░░ ECU · Governor · Fuel Injection
Dry-dock Operations ████████░░ Inspection · Repair · Reporting
Operation of Ship Power Plants — 4-year diploma program · graduated 2022. Thermodynamics · marine diesel engines · steam turbines · auxiliary machinery · ship electrical systems · automation & control.
STCW International: ISPS · Basic Safety Training (fire prevention, survival, personal safety) · Proficiency in Medical First Aid · Security Awareness Training.
Languages:
Maritime / Marine Automation
- Marine Automation Engineer
- Vessel Performance / Condition Monitoring Engineer
- Embedded Systems Engineer (Maritime)
- Naval Systems Integration Engineer
Hardware / ML systems (marine-domain expertise as differentiator)
- Hardware/Software Co-design Engineer
- FPGA / ASIC Engineer (LLM inference acceleration)
- ML Inference Optimization Engineer
- Pre-silicon Verification Engineer
"Most software engineers have never touched a marine diesel.
Most marine engineers have never written a GPU kernel.
Most GPU engineers have never designed the silicon underneath.
Three layers. One engineer."