piighost

piighost-proofreader

An LLM CV proofreader that anonymizes documents before any LLM call.

Repository Live demo

What it does

piighost-proofreader is an LLM-powered proofreader for CVs. You upload a PDF and get an annotated list of mistakes with click-to-highlight on the rendered pages. The document is anonymized with piighost-api before any LLM call.

How it works

  1. opendataloader-pdf converts the PDF to Markdown for the LLM.
  2. PyMuPDF renders each page and emits per-word bounding boxes.
  3. piighost-api anonymizes the Markdown before the LLM sees it.
  4. A LangChain and LiteLLM chain runs structured-output proofreading.
  5. A locator re-anchors each mistake to a page and bounding box.
  6. Streamlit renders the pages with overlays; clicking a mistake highlights it.

Run it

uv sync --group dev
cp .env.example .env  # fill in LITELLM_API_KEY etc.
uv run streamlit run app.py

You also need a running piighost-api at the URL declared in your .env.