I'm Eric, a physics graduate who likes to turn mathematics and physics into software. While I'd love to dive into the physics here, this space is dedicated to show the open-source software tools I build and maintain.

Originally built as proofs of concept while exploring architectural patterns and domain-driven design (DDD), these libraries have evolved into utilities for development. Published on PyPI.

• PyDepends: A lightweight, dependency injection framework designed for ease of use.
• PyMsgbus: An event-driven messaging framework for pub/sub architectures. (Read the Documentation)
• TorchSystem: A modular toolkit built to cleanly separate logic from messy infrastructure in PyTorch. (Read the Documentation)

Currently working on the Tannic Framework: a pure C++ machine learning ecosystem.

The ML landscape moves so fast that mainstream frameworks have become bloated with compilers, domain-specific sub-languages, and hardcoded CUDA kernels in strings. They often end up harder to use and maintain than plain C++ codebases. Furthermore, Python's interpreter overhead and memory unpredictability make deployment a nightmare. The bottleneck in production is no longer writing code, it's fighting Python.

Tannic is my attempt to address this. By moving machine learning logic back into C++ the goal is to reduce these friction points through a predictable inference ecosystem built from scratch.

• Tannic: The core C++23 tensor library. I am currently rewriting its execution model in C++26 to address some design flaws and further optimize performance and memory predictability. (Read the Documentation)
• Tannic-NN: The neural network inference library. See it in action through these implementation examples:
  • ConvNet on MNIST: A classic image classification model demonstrating the framework's 2D convolutional operations.
  • Vision Transformer (ViT): Show the framework's ability to handle modern transformer architectures.
  • Llama 3 Inference Server: A LLM inference server built with POSIX sockets to run Llama 3 models locally.
• PyTannic: Python utilities bridging PyTorch and Tannic via network communication, including tools to serialize PyTorch models into Tannic's custom format. (Read the Documentation)

If you are interested in my work, feel free to contact me.

• Email: eric.m.cardozo@gmail.com
• LinkedIn: eric-c