digital-registrar-gui 0.2.0b3

Streamlit inference GUI for the Digital Registrar pipeline.

digital-registrar-gui

Streamlit inference GUI for the Digital Registrar pipeline. Lets you paste a pathology report (or pick a folder of .txt files) and see the structured extraction in real time. Requires a local Ollama server (with one of the paper-benchmarked models) or an OPENAI_API_KEY — see Prerequisites below.

Install

pip install digital-registrar-gui
registrar-infer-gui

The default port is 8502. Override with registrar-infer-gui --port 9000.

Prerequisites — pick one LLM backend

The GUI is BYO-LLM. You need either:

• Local Ollama with one of the three paper-benchmarked models:

  ollama pull gpt-oss:20b      # default
  ollama pull qwen3:30b        # alt (Qwen3-30B-A3B)
  ollama pull gemma3:27b       # alt
  

• OpenAI: export OPENAI_API_KEY=sk-..., then pick gpt5_4_mini in the GUI's model dropdown.

See the main repo for the full pipeline docs and four-package overview.

What it does

• Two-column layout: input on the left, sticky JSON output on the right.
• Lets you switch engine (factory v2 or legacy v1), model, decomposition strategy, and optional jsonize / output validation.
• Folder mode: pick a directory, see one row per .txt, click to preview.
• Expander shows the DSPy LM trace for each run (router + group extractors).

Depends on digital-registrar for the actual pipeline.