Hi, I'm Mohamed Soubhi

13+ years architecting safety-critical automotive software at Valeo, TTTech Auto (TrustMotion), and Concentrico.
Bridging AUTOSAR & Modern C++ embedded platforms with Edge AI / TinyML and Graph Data Science.

engineer:
  name: Mohamed Soubhi
  role: Senior Automotive & Embedded Systems Software Engineer
  experience: "13+ Years (Valeo, TTTech Auto, Concentrico)"
  location: "Madrid, Spain 🇪🇸"
  core_domains:
    - AUTOSAR Classic (BSW, RTE, ASW) & Modern C++ (17/20)
    - Safety-Critical Architectures (ISO 26262 ASIL B/D, ASPICE, MISRA)
    - Edge AI & TinyML (TFLite Micro, INT8 Quantization, MCU Inference)
    - Graph Data Science & AI Agents (Neo4j GDS, GraphSAGE GNN, LangGraph)
  current_focus: "Zero-overhead embedded C++, on-device machine learning & automotive data platforms"

A technical conference presentation, interactive telemetry laboratory, and architectural reference on running real-time deep learning inference directly on resource-constrained microcontrollers without cloud reliance:

• 🖥️ Interactive 32-Slide Deck — complete conference talk with custom canvas scaling, animations, speaker notes, and embedded telemetry
• 🔬 Multi-Sensor Fusion Explorer — bidirectional simulation testing IMU vibration, acoustic, and environmental data classification
• 🔄 End-to-End Pipeline Blueprint — raw sensor acquisition ➔ DSP/MFCC feature extraction ➔ TensorFlow Lite Micro INT8 quantization ➔ bare-metal C++ execution
• ⚙️ The 4 Embedded Constraints — engineering for sub-milliwatt power budgets, kilobyte SRAM memory ceilings, < 10 ms latency, and tiny physical form factors

👉 Launch Live Project Hub · 🖥️ Open Slide Deck · 📂 View Repository

EU heavy-duty vehicle (truck/bus) powertrain benchmarking and $CO_2$ simulation platform built on European Environment Agency (EEA / VECTO) regulatory datasets (750k+ records):

• ⛏️ Automated Data Mining & Ingestion — SQL-over-HTTP pipeline mining live EEA Discodata, physical bounds validation (Pydantic), and high-performance column store (DuckDB + Apache Parquet)
• 🧠 ML Surrogate Modeling — Non-linear HistGradientBoostingRegressor (5-Fold CV, MAE 27.8) predicting certified $CO_{2v}$ emissions in sub-milliseconds from macro engineering specs
• 🛡️ Target Leakage Guard — Strict engine dyno cycle test isolation (assert_no_leakage) and SHA-256 data manifest tracking for regulatory auditability
• 📊 Interactive Multi-Tab Dashboard — OEM competitive powertrain analysis, component trade-off matrix, technology adoption curves, and real-time interactive $CO_2$ simulation

👉 Explore Live Demo · 📖 Read Technical Docs · 🖥️ Open Slide Deck · 📂 View Repository

An interactive web companion covering 116 course projects across 12 sections of Modern C++ (C++11/14/17/20), translated into bare-metal embedded systems architectures:

• 💻 Annotated Source Code — full syntax-highlighted viewers with multi-file tabs
• ⚡ Embedded Hardware Realities — ARM Cortex-M AAPCS calling conventions, instruction pipelines, struct padding, LittleFS, DMA framebuffers, MMIO wrappers
• 💡 Zero-Overhead Refactoring — production-ready, deterministic, MISRA-compliant modern C++ design patterns
• 🧠 464 Interactive Quizzes — instant self-checking questions with technical explanations

👉 Explore the Live Portal · 📂 View Source Code

End-to-end fraud detection knowledge graph on PaySim synthetic transactions (50k rows). Neo4j 5 + GDS 2.13 · GraphSAGE GNN · LangChain NL→Cypher · Ollama Cloud (deepseek-v4-flash):

• 🕸️ GDS Algorithm Pipeline — Account→Account graph projection with Louvain (community), PageRank, WCC, Betweenness, and Cypher cycle detection
• 🤖 Graph Neural Network Layer — 3-layer GraphSAGE trained on topological graph properties writing fraudProb ∈ [0,1] to every account
• ⚡ Sub-Second Performance — WCC in 20ms, sampled Betweenness 142× faster than exact, PageRank convergence in 2 iterations

👉 Explore Live Fraud Intelligence App · 📂 View Repository

Interactive machine learning and market analytics platform trained on over 100,000+ real vehicle listings to predict fair market valuation, non-linear depreciation curves, horsepower premiums, and options bundle lifts:

• 🏎️ Live In-Browser ML Valuation — instantaneous vehicle pricing adjusting for brand, body, mileage, horsepower, options, and accident history
• 📉 Dynamic Depreciation Decay — real-time 8-year resale value curves reflecting non-linear brand retention rates
• 💡 Value Attribution Waterfall — visual decomposition of base MSRP, power lift, package premiums, and age/mileage penalties

👉 Launch Live Valuation App · 📂 View Repository