_{llms.txt: an index of my work for language models · CV: English (PDF), 日本語・簡易版 (PDF)}

And what is good, Phaedrus,
And what is not good—
Need we ask anyone to tell us these things?

_{Robert M. Pirsig, Zen and the Art of Motorcycle code Maintenance (1974), the epigraph on its opening page}

think

write

• wave - feat. Federal Univ. of Santa Catarina and Ph.D P.V. Guimaraes
• math - pprint arXiv soon
• open & clean water - 6 BR authors in 2 countries
• open-science
• follow
• no AI, personal, wip

read

have fun

Staff Software Engineer & FDE 31yo

Piabrota AI Studio (self-employed, since October 2026)

• marola.piabrota.art: marola v2 in Haskell and other cool stuff, G R E A T news soon
• wave.piabrota.art: wave forecast in GPUs, >3k views on Zenodo, fresh on aiXiv
• piabrota: home
• room.piabrota.art: PC only, a 3D room - three.js, 3d model, ambience. idea
• metrics.piabrota.art: homelab numbers - vcpu / ram
• grid.piabrota.art: unstructured grids of the Brazilian coast - wind/temp now, swell later
• skills.piabrota.art: my skill collection - DYOR
• Partnerships or white-label requests: whitelabel [at] piabrota.art

Technical skills and background

• Software development: 10+ years. Mostly Scala for backend and data; Python for ML and Airflow.
• 2015: LIOC, ocean instrumentation lab. FPGA signal processing in LabVIEW (low code) for Brazilian Navy underwater communication
• 2016: Real-time GPS boat tracking over 2G, with mobile phone chips
• 2017: StoneCo, during its IPO. Churn prediction with XGBoost, to replace Bayesian networks
• 2018: a model that recommends a mobile carrier from ping history. Flink - during alpha stage - detects events; Python scores
• 2019: a pre-seed startup. We made an instant transfer between banks. It launched in August 2020, before Pix. A company bought us; then Pix made the product unnecessary
• [UK] 2020: an early startup based in London, in overseas banking. I worked mainly on the Scala backend and the FX APIs
• [US] 2021: a California startup. Semantix bought it
• [US] 2022: promoted to Lead MLOps, consulting for Americanas (B2W, Brazil). ML environments with Kubernetes and Kubeflow
• [US] 2023 to 2024: fraud prevention on Databricks. L5 engineer and Subject Matter Expert of a loss forecasting product. I reported to the CFO. "Values VIP: Tenacious" award, February 2024
• [UK] 2025: ML/MLOps in TV, rearchitect audience timeseries forecasts for 4 channels, >100 models run by day, indirect contract w/ TXP/UK. SageMaker, MWAA, S3, MLflow, XGBoost and symbolic AI
• [UK] 2026: I redesigned, with DS Lead, all 34 long-term models (2 channels × 17 demographics) as hierarchical heuristic rules. I took them to production on AWS. I resigned on early September 2026

open2work mhoffmannfs [at] gmail (.com)