ATR Workflow GSA

A semi-automated text recognition workflow for digital editions at the Goethe- und Schiller-Archiv (Klassik Stiftung Weimar). It combines Transkribus HTR with multimodal LLMs (Claude, GPT, Gemini) and generates TEI-XML output.

Note: This tool is primarily designed for use within the Goethe- und Schiller-Archiv. The two built-in project modes (Briefe an Goethe and Goethes Lyrik) include project-specific prompts and configurations. Adapting it to other projects requires changing project names, prompts, and potentially the TEI schema logic in the source code.

This project was developed with the assistance of Claude Code and GitHub Copilot.

---

Requirements

• Python 3.11+
• API keys for: Anthropic, OpenAI or Google Gemini
• Transkribus credentials

---

Installation

1. Clone the repository and create a virtual environment:

git clone https://github.com/KlassikStiftungWeimar/atr-workflow-gsa-public.git
cd atr-workflow-gsa
python -m venv venv
# Windows:
venv\Scripts\activate
# Linux:
source venv/bin/activate

2. Install dependencies:

pip install -r requirements.txt

3. Create a .env file in the ocr_workflow/ directory:

SECRET_KEY=your-django-secret-key
ALLOWED_HOSTS=127.0.0.1
# Optional: Required when running behind a proxy or with different domain
# CSRF_TRUSTED_ORIGINS=https://yourdomain.com

# API Keys – one set per project (RA = Briefe an Goethe, GL = Goethes Lyrik)
ANTHROPIC_KEY_RA=...
ANTHROPIC_KEY_GL=...
OPENAI_KEY_RA=...
OPENAI_KEY_GL=...
GOOGLE_KEY_RA=...
GOOGLE_KEY_GL=...

# Transkribus credentials
USERNAME_TRANSKRIBUS_RA=...
PASSWORD_TRANSKRIBUS_RA=...
USERNAME_TRANSKRIBUS_GL=...
PASSWORD_TRANSKRIBUS_GL=...

You only need keys for the providers you intend to use.

4. Run migrations and start the server:

cd ocr_workflow
python manage.py runserver

The app is available at http://127.0.0.1:8000/atr_workflow/.

Note (DEBUG=False): When running with DEBUG = False in settings.py (e.g. for production), Django no longer serves static files automatically. You must collect them first:

cd ocr_workflow
python manage.py collectstatic

This copies all static files into the staticfiles/ directory defined by STATIC_ROOT. You then need a web server (e.g. nginx or whitenoise) to serve that directory.

---

Usage

1. Select project mode

Toggle between Briefe an Goethe (RA) and Goethes Lyrik (GL) at the top. This controls which API credentials are used and which default TEI prompt is applied.

2. Upload images

Upload up to 10 document pages (JPG/PNG). All pages are processed together in one run.

3. Configure text recognition

Transkribus model – The HTR/OCR model used on Transkribus:

• The Text Titan I – General-purpose handwriting model
• The Text Titan I ter – Updated variant of Text Titan I
• German Genius – Optimized for German handwriting
• Transkribus Print M1 – For printed text

Multimodal LLM – The language model that independently reads the image and refines the result:

• Claude Sonnet 4.5
• Gemini 2.5 Pro
• Gemini 3 Pro Preview
• GPT-5.2

Temperature – Controls how deterministic the LLM output is (0.0–1.0). Lower values produce more consistent, conservative results; higher values allow more variation. Does not affect Transkribus.

4. Start recognition

Click Texterkennung starten. Transkribus and the LLM run in parallel. A merged result combining both outputs is generated automatically. Progress is shown in real time.

Once complete, the comparison view lets you display any two versions side by side:

• Transkribus – Raw Transkribus output
• Multimodales Sprachmodell – Raw LLM output
• Transkribus + Multimodales Sprachmodell – Merged result
• Eigene Version I / II – Editable fields for manual corrections

5. Generate TEI-XML

Switch to the TEI panel and configure:

Text source – Which text version to use as the basis for TEI generation:

• Eigene Version I / II – Use one of the manually edited versions (recommended)
• Transkribus + Multimodales Sprachmodell – Use the merged result directly

Prompt type – The instruction given to the LLM for TEI markup:

• Prompt Briefe an Goethe – Pre-configured prompt for the RA project
• Prompt Goethes Lyrik – Pre-configured prompt for the GL project
• Eigenes Prompt – Enter a fully custom prompt (image and recognized text are appended automatically)

LLM and temperature – Same options as in the recognition step.

Click Erstelle TEI-XML to generate. The result is shown alongside a plain-text comparison view.