rdiaz.dev
ricardo díaz·software engineer·machine learning
#selected work — 4 entries
  • 2025–26
    flash-sam3
    Replaced SAM3's 815M-param ViT-H encoder with TinyViT via knowledge distillation. 19× CPU speedup, 8.5× GPU speedup at 75% IoU retention.
    mlpytorchpython
  • 2025–26
    flash-attn-fpga
    Tiled FlashAttention IP core in Vitis HLS for FPGA/SoC platforms. Eliminates full N×N attention matrix materialization; seven pipelined modules via AXI4.
    fpgahlsc++
  • 2025
    maguito
    Terminal-based Git interface in Rust inspired by Magit. Hunk-level staging without spawning Git subprocesses; collapsible section tree for staged/unstaged changes.
    rusttuigit
  • 2017–24
    ruby-on-rails
    26 commits to the Rails framework across Action Dispatch, Active Support, Active Record, and Active Storage, ranging from test infrastructure to API cleanup. Ranked #273 among all-time contributors.
    rubyopen-source
bufprojects·themedark