A short shelf of public work

Start with the problem.
Follow the work.

LazyingArt builds inspectable tools around books, language, media, research, and local agents. This is a short shelf, not a catalogue of every experiment: open the source, read the output, or start with one small piece of work.

CODETools you can inspect
BOOKSAligned reading editions
MEDIAText tied to time
RESEARCHRebuildable evidence

Selected technical writing

Explain the decision.
Show the work.

These articles begin with a practical problem, walk through the tradeoffs, and leave readers with commands, evidence, or a reproducible route they can inspect.

Developer tools · first-person build story

I got tired of agent limits, so I built AgInTiFlow

How a collection of scripts became a reviewable workflow for long-running coding agents, with the failures and operational boundaries left visible.

Read the article →

Private knowledge · practical guide

Search confidential PDFs locally without overbuilding RAG

A small decision path for extraction, local search, citations, and the point where a heavier retrieval system becomes justified.

Read the guide →

LaTeX · reproducible workflow

Build a LaTeX redline that still compiles

A baseline-first workflow for latexdiff, generated-file repair, three clean builds, and a change trail that survives editorial review.

Read the guide →

Books and local knowledge

Find the passage,
not just an answer.

Private libraries and multilingual books become much more useful when every card, note, and translation can lead back to its source.

Local knowledge · code

Local Knowledge Terminal

Builds cited multilingual cards from a collection while keeping the books, indexes, and model on local hardware.

Inspect the terminal →

Multilingual books · code

PocketPolyglot

Turns aligned text into compact editions with pinyin, furigana, grammar colour, editable source, and reproducible PDF output.

Inspect the book pipeline →
Need to test your own collection? See the Local Knowledge Terminal sprint →

Lecture and study media

Make one recording
worth returning to.

A useful lecture pack keeps the transcript, translation, subtitles, and notes on one timed source instead of letting each become a separate copy.

Long-form media · code

Video2Book

A pipeline from media and timestamped transcripts to structured notes, subtitles, PDFs, EPUBs, and pocket editions.

Inspect Video2Book →

Captions and clips · code

LazyEdit

Keeps transcription, subtitle translation, review copies, caption burn-in, and publishing preparation in one editable workflow.

Inspect LazyEdit →

Subtitles + study pack · complete sample

One source, every promised file

Open a project-owned English source, corrected timeline, Traditional Chinese alignment, SRT tracks, A5 study PDF, captioned preview, issue ledger, and rights manifest.

Inspect the complete delivery →
Watch three selected video pieces and see the production role → Have one rights-cleared lecture? See the Bilingual Lecture Pack →

Research and manuscripts

Keep the clean paper
and the change trail.

Revision is safer when the baseline, current manuscript, generated redline, commands, logs, and open issues remain separate and rebuildable.

Revision workflow · agent skill

Paper revision skill

A plan-gated workflow for LaTeX revisions, response letters, proof corrections, build checks, and visible unresolved issues.

Inspect the workflow →

Redline · reproducible sample

Three-build LaTeX proof

Baseline, clean revision, generated redline, final logs, pinned tooling, and hashes in a small synthetic example.

Inspect the three builds →

Local workspace · code

PaperAgent

Edit LaTeX and code in the browser, run Python or R, compile on the backend, and inspect the PDF beside its build log.

Inspect PaperAgent →

Qualitative coding · reproducible sample

Auditable content coding

Freeze the codebook, keep each choice beside its exact excerpt and locator, flag ambiguity, and rebuild the same report from a hashed manifest.

Inspect the synthetic sample →
Need one bounded manuscript repaired? See the Build & Redline Sprint →

Agents and instruments

Give an agent
something real to operate.

Useful agents need bounded tools, visible state, recoverable actions, and outputs that survive the chat that created them.

Reusable workflows · code

LazySkills

Portable working instructions for books, publishing, media, browser operation, research, hardware, and other long-running tasks.

Browse the skills →

Scientific design · code

AgInTi LabCanvas

An agentic studio for editable scientific figures, CAD devices, and experiment design where the artifact remains open to revision.

Inspect LabCanvas →

Provenance · offline code

Proofline

Checks local file hashes, transformation edges, and source-located claim evidence without uploading the underlying material.

Inspect the passing example →

Seven small ways to begin

Bring one bounded problem.

The selected fixed scopes are USD 250 or USD 500. Each starts with a free fit check or scoped inquiry; source transfer and payment come only after the exact scope is accepted.

01

Local Knowledge Terminal

Test one private collection, one language goal, and one existing machine.

See the sprint →
02

OpenHI Software Reproducibility · USD 500

Run one named public software stage on one existing workstation and rights-cleared dataset, with the environment, command, output, failures, and go/no-go kept together.

See the executed sample and scope →
03

Book Specimen Sprint

Test one rights-cleared chapter in print-layout PDF and reflowable EPUB before producing the whole book.

See the sprint →
04

Bilingual Lecture Pack

Turn one short, rights-cleared English lecture into linked study material.

See the pack →
05

Manuscript Build & Redline

Repair one bounded LaTeX build and leave a reproducible change trail.

See the sprint →
06

Story Clip Pilot

Find two candidate moments in one owned recording, then finish the buyer's selected vertical clip.

See the pilot →
07

KiCad Plugin Evaluation · USD 400

Run one authorized geometry-plugin version through four agreed presets on a clean seven-case fixture, with before-and-after evidence, DRC comparison, and an issue report.

See the fixture and scope →