Skip to content
View wolfwdavid's full-sized avatar

Highlights

  • Pro

Block or report wolfwdavid

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
wolfwdavid/README.md

David White Wolf

ME x ML Engineer — Mechanical engineer building production ML systems for real engineering and financial problems.

Most profiles are CS grads doing sentiment analysis on IMDB. I build ML that solves actual engineering problems — because I've spent years doing the engineering first.


What I Build

Physics-Informed Deep Learning — Neural surrogates that predict structural analysis 1000x faster than FEA. Deep ensembles with calibrated uncertainty. Loss functions that encode beam theory, not just MSE.

Domain-Adapted LLMs — Fine-tuned Qwen2.5-3B on 5K+ mechanical engineering Q&A pairs. Custom MechEval benchmark proves the 3B model outperforms the base 7B on domain tasks.

Agentic AI Systems — Custom ReAct agent with 7 tools for market microstructure analysis. Hand-built loop (no LangChain), two-tier cache, provider failover, rate limiting.


Pinned Projects

Project What It Does Stack Demo
fea-surrogate Physics-informed neural surrogate for structural analysis PyTorch, Gradio, Docker, CI/CD Live
mechspec-qwen Domain-adapted LLM for engineering specs and calculations Transformers, PEFT/LoRA, Gradio Live
agentic-market-analyzer Custom ReAct agent for market analysis — no frameworks Qwen3-8B, yfinance, FRED, Gradio Live

Technical Stack

Python PyTorch Hugging Face Docker React GitHub Actions TensorFlow Node.js

ML/AI: PyTorch, Transformers, PEFT/QLoRA, Gradio, Deep Ensembles, Physics-Informed Loss Engineering: Euler-Bernoulli, Kirchhoff-Love, Lame equations, FEA/FEM, GD&T Infrastructure: Docker (multi-stage), GitHub Actions, MLflow, model versioning Full-Stack: React, Node.js, WebSockets, D3.js, SQLite, Svelte


Background

  • Mechanical engineer (ME) — structural analysis, FEA/FEM, GD&T
  • ME + ML crossover: I understand both the physics and the models
  • Active trader — built Monte Carlo options analyzer, real-time market terminal
  • NYU engineering
  • 203 tests across 3 ML projects, all passing

Pinned Loading

  1. agentic-market-analyzer agentic-market-analyzer Public

    Custom ReAct agent for market microstructure analysis — 7 tools, no LangChain

    Python

  2. fea-surrogate fea-surrogate Public

    Physics-informed neural surrogate for structural analysis — 1000x faster than FEA with >99.9% accuracy

    Python

  3. fraud-detection-tf fraud-detection-tf Public

    Production fraud detection with TensorFlow, Vertex AI pipeline, FastAPI serving, and model monitoring

    Python

  4. jax-pinn jax-pinn Public

    JAX/Flax physics-informed neural network with jax2tf export — benchmark JAX vs PyTorch vs TensorFlow

    Python

  5. mechspec-qwen mechspec-qwen Public

    Domain-adapted Qwen2.5-3B for mechanical engineering specs, materials, and GD&T

    Python

  6. vision-edge vision-edge Public

    MobileNetV3 object detection with TFLite quantization — fp32/fp16/int8 edge deployment benchmarks

    Python