Skip to content
View h0ffmann's full-sized avatar
🏊
producing bugs
🏊
producing bugs

Organizations

@iu-pay

Block or report h0ffmann

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.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

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

Report abuse
h0ffmann/README.md

📄 cv (en) · 📄 cv (pt-BR) · 📄 cv (ja) · 🇧🇷 português · 🇯🇵 日本語

👋 me

sr. software engineer & jr. researcher with 8 years of experience building scalable (and occasionally clever) solutions mainly for payment processing & fraud-fighting, now building open-source ocean tech at marola-dev

LinkedIn

🌊 open source — marola-dev

QR code for marola.dev

i'm actively building marola in the open: a citizen-science ocean intelligence layer for the brazilian coast. it answers "can i swim tomorrow, and when?" for every beach around florianópolis, rio de janeiro and salvador: the best hour from swell, wind, tide, uv and the official bathing-water bulletins, with the why in plain language, live at marola.dev.

  • water quality first - every official sampling point (ima/sc, inea, inema) is on the map, coloured by its last bulletin; unfit water zeroes the score in deterministic scala 3 (kyo) code, and the llm explains but never overrules
  • the sea in front of you - tide, swell, wind, uv, jellyfish and whale odds per beach, a sourced did you know? about the local sea, and ask the ocean: rag with citations on ollama, an mcp tool server and its own small model, marola-sea, no account or api key
  • designed in public - every non-trivial change starts as a public design doc (MIPs); next up: wind, wave, satellite and sea-heat map layers, ressaca (storm surf) warnings, wave-model ensembles with WAVEWATCH III and an open data lake of beaches and water quality

marola.dev github: marola-dev docs.marola.dev MIT license DOI for marola

last commit to marola-dev/marola monthly commits to marola-dev/marola marola CI marola.dev build + deploy

repo what
marola the umbrella: design docs, roadmap and docs.marola.dev, every repo below as a submodule
marola-app the product: scala 3 + kyo pipeline, score and safety veto, cli, mcp tool server
marola-site the map at marola.dev, rebuilt every 3 hours
marola-corpus the sourced ocean knowledge every answer cites
marola-ml offline python: dspy prompt compile, benchmark gate, the marola-sea model
marola-oods open ocean data store: an open data lake of beaches and bathing-water samples (in progress)
marola-devkit shared dev harness: nix tools, hooks, claude code skills, ci workflows
awesome-ocean-science curated list of open ocean models, data, tools and institutes
open-sustainable-technology fork of the community list of open sustainability projects
awesome-open-climate-science fork of the community list of open climate science

civic-tech disaster-risk-management forecasting meteorology non-profit oceanography water-quality

🔓 contributors welcome, humans and ai agents alike — no code needed: use the map and tell us where it's wrong, share local sea knowledge, or open an issue or a PR at marola-dev/marola. or mail me: mhoffmannfs[at]gmail.com

⚡ research — ww3-gpu

my undergraduate project at ufrj (escola politécnica), co-advised at labeco/ufsc: ww3 gpu lab, an open lab for running noaa's WAVEWATCH III wave model and moving its expensive kernels to gpus without changing the answer. a nix-pinned fortran/mpi/netcdf toolchain builds ww3 and runs a regression test in one command. the dia nonlinear-interaction kernel (W3SNL1), ported to c++/kokkos, matches the fortran bit for bit on serial, openmp and cuda, and runs 1,000 sea points in 0.047 ms on an rtx 4090 against 24.96 ms serial. it also holds a 16-lesson course, from the wave spectrum to porting kernels with coding agents.

github: ww3-gpu DOI for ww3-gpu

💼 exp-highlights

  • itv (uk) - ml/mlops on aws for channel audience forecasting (2025/aug 2026)
  • signifyd (us - award winner tenacious feb/2024) - ml eng. for loss forecasting & risk-analysis on databricks (2023/2024)
  • elemeno AI (us - br client bvmf: AMER3) - mlops engineer with k8s on gcp (2021/2022)
  • broad (uk - YC W21) - sr. software engineer with scala for digital banking on gcp (2020)
  • stone (br: nasdaq: STNE) - ml & software engineer with scala on aws (2017-2019)
  • LIOC (br: UFRJ/COPPE) - ocean instrumentation lab - scientific research, ROV, microcontrollers (2016/2017)

🏅 certs

AWS Cloud Practitioner AWS ML Specialty AWS ML Engineer Associate AWS Data Engineer Associate

📚 learning

studying for Google Cloud Professional Agentic Architect

🏊 misc

strava (open water swimming)

PADI Advanced Open Water Diver PADI Enriched Air (Nitrox) Diver PADI Rescue Diver

Pinned Loading

  1. marola-dev/marola marola-dev/marola Public

    ocean intelligence layer

    Python 15 5

  2. ww3-gpu ww3-gpu Public

    WW3 GPU Lab: agentic coding and agentic research on a bit-for-bit GPU port of WAVEWATCH III

    C++

  3. swim-sync swim-sync Public

    swim training app

    TypeScript

  4. montaigne montaigne Public

    next level text-to-sound

    Python

  5. piabrota/ihoje piabrota/ihoje Public

    'simple' event scrapper

    Rust

  6. okekara okekara Public

    synchronized lyrics and karaoke mode to YouTube Music

    TypeScript