§ 01 fig. 1.1
DATA
Modeling, engineering & analytics. Pipelines, dimensional modeling, lakehouse, ETL/ELT, data quality. From raw ingestion to the table the business team trusts.
› // warehouse · lakehouse · streaming · dq
vol. I · iss. 01 · 2026 · são paulo / br
We model
data,
not opinions.
A lab building data engineering, machine learning and platform — for companies that need fast technical resolution and architecture that survives the next scale. What we learn along the way, we publish: open source, devlogs, and in-house products.
· data engineering
· machine learning
· platform
· open source
§ 01 § 01 · what we build
Four fronts — one obsession with architecture that holds.
§ 02 fig. 1.2
MACHINE LEARNING
Models that ship to production. Classical ML, deep learning, computer vision. Real MLOps — model versioning, continuous eval, drift monitoring.
› // classical ml · deep learning · cv · mlops
§ 03 fig. 1.3
PLATFORM
Functional full-stack. Frontends rebranded or rebuilt from scratch, automations that replace operational work, internal platforms with observability from day one. QML when the case calls for it.
› // web · automation · qml · rebrand
§ 04 fig. 1.4
OPEN SOURCE
Most of what we build is public. Every paid project becomes a library, tool, or paper. You get what we learned along the way — and the next client arrives knowing what they’re walking into.
› // libs · clis · tools · mit
§ 02 § 02 · apps · coslu
Apps · COSLU
Open source with the COSLU brand baked into the UI. The technical showcase of what the lab ships when we’re our own client.
№ 001 MIT
MARKDOWNREADER
100% local Markdown editor. No cloud, no login, no 200MB of Electron.
- 3.8MB binary · Tauri + React + CodeMirror 6
- Split-pane · GitHub-flavored Markdown · syntax highlighting via Shiki
- Local vault — open a folder, index, edit
№ 002 MIT
JARVSTRANSCRIPT
Real-time transcription on your PC, GPU accelerated.
- Real-time STT · CUDA · Windows 11
- Global hotkey F8 — start transcribing from anywhere
- 100% local · audio never leaves your machine
№ 003 MIT
COSLU READER
Universal file viewer. 80+ formats in a 12 MB binary.
- PDF · DOCX · XLSX · PSD · Parquet · SQLite · glTF · EPUB
- On-demand loading · O(1) cold start on catalog size
- Fallback chain: viewer → text → hex · nothing ends in error
CL
EST · 2026
§ 03 · manifesto
¶ We build in public.
The lab exists because the best way to prove we can build is by building. Every service we sell becomes a library, devlog, tool or product — so the next client arrives with auditable references, not a PowerPoint case study.
CODE
Everything we write is versioned and auditable.
OPEN SOURCE
Most of it is public. MIT whenever possible.
LOGIC
Decisions based on metrics, not opinion.
UTILITY
If it doesn’t solve a real client pain, we don’t build it.
§ 04 · contact
Got a data or platform problem no one has helped you solve?
Send a brief. Within 48h we reply with scope, timeline, and what would be open-sourced along the way.