Good point! Think of it as a single trusted reference you bookmark
instead of searching each time and landing on random sites with ads
and cookie banners.
Beyond calculators, there are tools you won't find easily elsewhere:
AI background removal and depth estimation running entirely in your
browser via WebGPU — no uploads, 100% private. More coming.
The goal is to make Flowelio a portal for bleeding-edge client-side
tech, not just a calculator site
I've been building Flowelio, a growing suite of free calculators
and tools with one hard constraint: everything runs client-side.
No servers, no file uploads, no accounts — just the browser.
What's in it right now:
- Health: BMI, TDEE, body fat, pace calculator, 1RM, race predictor
- Finance: compound interest, ROI, break-even, carbon footprint, solar savings
- Utility: password generator, unit converter, timezone, percentage, split bill
- Italy-specific: net salary (with full IRPEF/INPS breakdown), flat tax regime,
NASpI unemployment, property tax (IMU) for all 8,100+ Italian municipalities,
car tax by region, fiscal code generator/validator
- AI tools: background removal and depth map estimation — both running
entirely in the browser via Transformers.js + WebGPU, zero uploads
The Italian tax tools were the most interesting engineering challenge.
Italy has 8,100+ municipalities, each with its own property tax (IMU) rate
set by local government. Rather than fetching this at runtime, I baked all
the data into static JSON files and used Angular's SSG to pre-render a
dedicated HTML page for every single municipality at build time.
The result: ~8,400 static pages deployed to Cloudflare's edge,
sub-50ms TTFB globally, zero infrastructure cost.
The privacy angle came naturally from the architecture — when there's
no server, there's nothing to leak. The AI tools make this concrete:
background removal competitors like remove.bg upload your photo to their
servers. Ours processes everything locally with ONNX Runtime + WebGPU.
More countries (US, UK, DE, FR, ES, CA, AU, IN, BR) and dev tools
(JSON formatter, regex tester, diff checker, SQL playground) are in progress.
Happy to discuss the SSG strategy, the tax data pipeline, or the
client-side AI approach. Would love to hear what tools you'd find
most useful.
I've been thinking about something that feels off in how we build AI.
Modern language models are impressive at generating text, yet they're fragile
in ways human cognition isn't. They lack persistent reasoning, genuine goals,
long-horizon planning.
The intuition: what if language isn't the foundation of thought, but just
an interface to it? What if reasoning happens in a latent, continuous,
non-linguistic space we rarely inspect?
We've optimized prediction of the next token. But token prediction isn't
cognition—it's an output modality. Researchers like Yann LeCun and Karl
Friston have been pointing at this for years: biological intelligence emerges
from continuous, predictive, embodied interaction with the world.
Our AI still treats language as the primary substrate of thought. Maybe that's
the bottleneck.
I've been exploring this informally—nothing validated, no claims of novelty,
just thinking out loud with some structure. I collected rough ideas, open
questions, and a sketch of what this might mean for architecture design.
Curious what HN thinks. Is this direction interesting? Obviously wrong? Am I
just redescribing what transformers already do?
Beyond calculators, there are tools you won't find easily elsewhere: AI background removal and depth estimation running entirely in your browser via WebGPU — no uploads, 100% private. More coming.
The goal is to make Flowelio a portal for bleeding-edge client-side tech, not just a calculator site