Ideas into things —

Build. Explore. Figure it out.

I’m Aduneer, and I love learning. I build things to understand them, then keep refining what I make and how I think. Every project is a chance to get a little better.

View my work How I keep learning

55 AA// still curious

DRAG / TAP TO SPIN

Selected projects

A few things I’ve built

Inside each project A real preview, the decisions behind it, and a direct way to try it.

01 Error monitoring

Faultwing

Self-hosted error monitoring service built in Go. Explore project

Why it exists Catch bugs before they infest production.

Faultwing dashboard with the issue terrarium, issue list, and selected error details — Faultwing’s local demo dashboard

From exception to issue

01App / SDK
02Go API
03Worker
04Issue

Built around: A Go API and worker, PostgreSQL, a React dashboard, and small Go, Python, and Node.js clients.
Includes: Error grouping, issue states, realtime dashboard refreshes, and stack-frame links for common editors.
Current scope: An early-stage learning project. The local workflow is complete; production deployment and security hardening are not.

Try local demo Source

02 Terminal learning

FlashTerm

A terminal study tool that makes you type the answer. Explore project

Why it exists No multiple choice and no self-grading: produce the answer, or keep working on it.

FlashTerm review showing a Spanish vocabulary card, typed answer, and correct result — A real review session · click to watch the demo

Five-box review schedule

Box 11 day
Box 23 days
Box 37 days
Box 414 days
Box 530 days

Built around: A C++17 terminal application using plain-text decks and no third-party dependencies.
Includes: Leitner scheduling, reverse review, typo tolerance, hints, undo, tags, and progress statistics.
Deck format: Custom CSV decks, cloze deletion, and optional image or audio support without a required service account.

Watch demo Source

03 Early experiment

Fugazi

A tiny language model that borrows the rhythm of a text corpus, one word at a time. Explore project

Why it exists An early step into AI and ML: seeing what a small, understandable text model can generate.

$ python -m fugazi --seed 1 --sentences 3
We built a platform. Platforms are wonderful because they can do anything.

Excerpt from a reproducible run on the bundled sample corpus

From corpus to sentence

01Corpus
02Tokens
03Contexts
04Next word

Mechanism: A word-level Markov chain that samples from the words observed after each context in the corpus.
Includes: Seeded repeatable runs, a cache keyed to corpus metadata, and clear errors for unusable input.
Scope: A classical n-gram experiment that runs on Python's standard library, with its limitations documented.

Run locally Source

More work FlatScorer Gruber Darker for Obsidian All repositories

Build small✦ Break it✦ Read the source✦ Follow the signal✦ Ask why✦ Keep learning✦ Try again✦ Make it better✦

Field notes

Decisions behind the demos

From question
to working system.

01 / Faultwing

Let the worker do the grouping.

An incoming exception becomes a durable PostgreSQL job. A Go worker fingerprints and groups it into an issue; workers can claim jobs concurrently with FOR UPDATE SKIP LOCKED. The dashboard refreshes from the authoritative API.

Read the architecture

02 / Fugazi

The chain remembers only its previous words.

Fugazi counts what follows each short context, then samples a path through those transitions. Starting at sentence openings and fixing the random seed make the output easier to read and reproduce. It was one of the experiments that drew me into language modeling.

Read how it works

Open signal cache 640 × 480 / 16 colors

build / learn / repeat

A small nod to Terry A. Davis and the handmade world of TempleOS (opens in a new tab).

the pixels glow in the dark