A collection of standalone tools and utilities, each in its own subdirectory with dedicated documentation. Ordered below from most broadly useful to most specific.
Python CLI for bulk filesystem operations. Three subcommands:
git— recursively finds Git repositories under a source tree and moves them to a centralized destination, with configurable conflict resolution (number,timestamp,uuid).cleanup— removes build artifacts and cache directories (__pycache__,node_modules,target,.venv,.DS_Store, and more) across a directory tree.organize— sorts files by type into categorized subdirectories (videos/,audio/,images/,documents/pdf/,code/python/, etc.), respecting a configurable depth limit.
All modes support --dry-run for safe previewing, --include-hidden, --cleanup-empty-dirs, and --verbose. Installable as a package via pyproject.toml.
uv run file-organizer git -s ~/scattered -d ~/organized --dry-run
uv run file-organizer cleanup -s ~/projects
uv run file-organizer organize -s ~/downloads -d ~/sorted --max-depth 2Technology: Python (Typer). See ./file_organizer/readme.md.
Two uv-runnable scripts for converting PDFs to Markdown, covering different tradeoffs:
pdf_to_md.py uses marker-pdf — ML-based layout analysis with GPU support. Best for complex documents: multi-column layouts, mixed text/figures, scanned pages. Supports --force-ocr and a --debug mode that saves layout images and bounding-box JSON.
pdf_to_md_light.py uses pymupdf4llm — no GPU, no large models, near-instant on native PDFs. OCR runs selectively only on pages that need it. Supports image extraction to disk or embedded as base64, selective page ranges, and Tesseract language configuration.
uv run pdf_to_md/pdf_to_md.py paper.pdf # marker-pdf, full ML
uv run pdf_to_md/pdf_to_md_light.py scan.pdf --force-ocr --lang por # pymupdf, fast
uv run pdf_to_md/pdf_to_md_light.py thesis.pdf --pages 0,5-10 --imagesTechnology: Python (uv scripts, marker-pdf / pymupdf4llm).
Downloads a YouTube video's transcript (auto-generated or manual) from a URL or bare video ID. Resolves all URL formats: youtu.be, /shorts/, /embed/, /live/, and standard ?v= links.
Output formats: plain text (default), SRT with timestamps, or raw JSON. Language priority falls back through pt-BR → pt → en by default, configurable per invocation. Output file name derived from the video ID when no explicit path is given.
uv run --with youtube-transcript-api youtube_transcripts/download_transcript.py https://youtu.be/dQw4w9WgXcQ
uv run --with youtube-transcript-api youtube_transcripts/download_transcript.py <url> -f srt -o sub.srt
uv run --with youtube-transcript-api youtube_transcripts/download_transcript.py <url> -l en -l ptTechnology: Python (youtube-transcript-api). See ./youtube_transcripts/requirements.txt.
Multi-threaded C++ CLI for converting text file encodings in-place. Handles any encoding pair supported by iconv, including the common case of latin1 files misidentified as UTF-8.
Key behaviors: atomic replacement via mkstemp + rename(2) (no partial writes), two strategies for unrepresentable bytes (drop the byte or remove the whole line), recursive directory processing parallelized across all CPU cores, and session resumption for interrupted directory runs.
meson setup build && meson compile -C build
build/file_cleaner ./legacy -s latin1 -t UTF-8 -r -f
build/file_cleaner document.txt -t latin1 -f --remove-invalid-lines -vTechnology: C++17 (Meson, iconv, mmap). See ./TextSanitize/readme.md.
Bash script that automates the full LaTeX compilation sequence, including the pdflatex → biber → pdflatex → pdflatex cycle needed for Biber bibliographies. Three modes: build and open the PDF (-b), monitor source files and recompile on change (-m), and clean build artifacts (-c). Includes dependency checks and installation hints for apt-based systems.
./latex_compile/latex_build.sh -b # compile main.tex and open
./latex_compile/latex_build.sh report.tex -m # watch mode
./latex_compile/latex_build.sh -c # clean build dirTechnology: Bash (pdflatex, biber, inotify-tools). See ./latex_compile/readme.md.
Python script for extracting text from image and video files via Tesseract. Handles grayscale conversion and adaptive thresholding as preprocessing, configurable language codes, and frame-interval sampling for video. Output goes to stdout or a file.
uv run ocr/main.py scan.png -l por+eng
uv run ocr/main.py lecture.mp4 -f 60 -o transcript.txt
uv run ocr/main.py image.jpg --no-preprocessTechnology: Python (pytesseract, opencv-python). See ./ocr/readme.md.
C++ utility that recursively scans a directory for image files containing camera EXIF metadata (make/model tags) and copies them to a destination directory, skipping files already present by filename. Reports progress and a summary of scanned/copied/skipped/error counts.
./photo_finder /mnt/card /home/user/photosTechnology: C++17 (Exiv2, CLI11, spdlog). See ./photo_finder_cpp/readme.md.
Python module wrapping the Helsinki-NLP OPUS-MT model for English-to-Portuguese neural translation. Handles sentence splitting and batch processing internally. Intended for import rather than direct invocation.
from translate import translate_english_to_portuguese
result = translate_english_to_portuguese("Hello world")Technology: Python (transformers). See ./translate/readme.md.
Two Python scripts for TOTP (RFC 6238) code generation from an otpauth:// URI:
main.py— standalone script with step-by-step debug output showing the full HMAC-SHA1 derivation. Useful for understanding or auditing the algorithm.lacis_code_telegram_bot.py— minimal Telegram bot that returns the current TOTP code on/code. Requires a bot API key and thepython-telegram-botlibrary.
Technology: Python (standard library + python-telegram-bot). See ./totp/readme.md.
C++ program that traverses a directory tree and reports total size and file count per extension, sorted by size, count, or both. Supports depth limiting and a top-N filter. Progress indicator on stderr during scan.
./sizes_cc/sizes ~/downloads --top 10 --mode both
./sizes_cc/sizes /var/log --depth 1Technology: C++17 (CLI11). See ./sizes_cc/readme.md.
GNOME Shell extension that adds a panel indicator tracking recently copied text. Items can be re-selected from the menu (which moves them back to the top of the list). History size and polling interval are configurable via the preferences dialog. Tracks text clipboard contents only.
make -C gnome/clipboard_history install
gnome-extensions enable clipboard-history@sidtools.lucas
gnome-extensions prefs clipboard-history@sidtools.lucasTechnology: JavaScript (GNOME Shell API, GSettings). See ./gnome/clipboard_history/README.md.
Two browser userscripts (Tampermonkey/Violentmonkey):
tampermonkey_chatgpt_temporary.js— appendstemporary-chat=trueto ChatGPT URLs, forcing all sessions into temporary mode.tampermonkey_regenerate_deepseek.js— auto-clicks the Regenerate button on DeepSeek Chat when a server-busy error is detected.
Install by pasting each script into the Tampermonkey editor. See ./tampermonkey_scripts/readme.md.
Python module wrapping lucas-leme/FinBERT-PT-BR (a BERT model fine-tuned on Brazilian Portuguese financial text) for positive/negative/neutral classification with confidence scores. Designed for import; works reasonably on general Portuguese text despite the financial training domain.
from sentiment_analysis.sentiment import analisar_sentimento
sentimento, confianca = analisar_sentimento("Este produto é excelente.")Technology: Python (transformers, torch). See ./sentiment_analysis/readme.md.
Python scripts for downloading Instagram stories using session cookie authentication. capture.py downloads stories for a given username to ./story/<username>/ and prints story metadata as JSON on stdout. speech_recon.py optionally transcribes audio from video stories via Google Speech Recognition.
python insta_stories_download/capture.py <username> --cookies '{"sessionid": "..."}'Technology: Python (requests, lxml, moviepy, speech_recognition). See ./insta_stories_download/readme.md.
PowerShell scripts for running lightweight HTTP file servers on Windows. Three variants covering basic directory listing (dllhost.ps1), range request support for resumable downloads (syswinmanager.ps1), and an enhanced version with chunked transfer (windllhost.ps1). All serve on port 3306 and require running as Administrator.
.\expl\syswinmanager.ps1
# access at http://server:3306/Technology: PowerShell. See ./expl/readme.md.