🔒 Your data never leaves this machine. The file is read and cleaned locally; your original is untouched — no upload, no API, no cloud.
Reading your file…

Fix the messy text in your spreadsheet.

Misspelled names, phone numbers and emails in a dozen formats, cities typed five different ways — I'll correct and standardize them, protect anything sensitive, and show you exactly what I changed. I never delete your data; every change is reversible, and the judgment calls are left to you.

What "cleaning" looks like — same rows, just fixed
nigeiaNigeria CalforniaCalifornia Ana@GMAIL.com ana@gmail.com 415.555.0192(415) 555-0192
It corrects and standardizes messy text and removes duplicate rows — it doesn't delete your data. New to this? watch it run on a sample file →

Drop your export here

CSV or Excel. I'll scan it for the usual mess — duplicates, blanks, mismatched formats, stray spellings — and anything that looks like personal data.

try it on a messy sales export or an HR file with sensitive data
↻ have a saved recipe? apply it to this file
Cleaning up — this runs locally, so it's quick.
file.csv
The summary, in plain English
    Handled — already applied (and reversible)
    Before → after preview
    The audit trail (no silent edits, ever)

    Your clean copy is ready

    Take the cleaned file and the change log. Both are yours to keep.

    Your original is untouched — every change above is a named, reversible operation. Save the recipe and re-apply this exact cleaning to next month's export in one click.
    ← Clean another file

    How this demo covers the challenge

    🏔️ Tiny Titan

    One 4-billion-param model plans every clean — nothing bigger, anywhere.

    🔌 Off the Grid

    No third-party AI APIs — a local-runnable GGUF does the work (on-device when self-hosted).

    🎛️ Well-Tuned

    Custom QLoRA fine-tune, trained on execution-verified data and published on the Hub.

    🎨 Off-Brand

    Hand-built gr.Server interface — zero default Gradio chrome.

    🦙 Llama Champion

    Served through llama.cpp as a Q8_0 GGUF.

    🤝 Sharing is Caring

    Every run's agent traces are published as an open dataset on the Hub.

    📓 Field Notes

    A full build report — the failures documented next to the wins.

    🏡 Backyard AI track

    A hands-off cleaner for the non-coder with a messy Monday export.

    You've made 0 decisions — apply them and re-clean.