Contrary to what I'm reading, #Claude is far worse than I remember. It often doesn't fully understand simple questions and needs nudging - and as in this case, can't solve even trivial problems which only require careful reading of the log.

It took #Kimi 10secs to solve a problem with an api call to a #docling
server crashing when accessed over its api. I used Kimi in the browser. #ClaudeCode Sonnet 4.6 had full access to the actual server and couldn't solve the issue after 30 mins of repeated attempts.

```
The root cause is clear: **Docling workers are crashing due to missing EasyOCR dependency**, causing the async jobs to fail. OpenWebUI polls for results, gets a 404 because the job died, and reports "Task result not found."

**Fix the Docling installation:**

```bash
# Stop the service
sudo launchctl unload /Library/LaunchDaemons/com.docling-serve.plist

# Install EasyOCR in the correct uv tool environment
uv tool install --upgrade docling-serve --with easyocr

# Or if that doesn't work, run the specific Python interpreter:
/Users/urisharf/.local/share/uv/tools/docling-serve/bin/python -m pip install easyocr

# Reload
sudo launchctl load /Library/LaunchDaemons/com.docling-serve.plist
```

**Alternative (if you don't need OCR):** Configure Docling to use a different OCR engine or disable it. Check Docling's env vars for `DOCLING_OCR_ENGINE` or similar.

**Verify the fix:**
```bash
curl -X POST http://mbp.lan:5001/v1/convert/source \
-H "Content-Type: application/json" \
-d '{"source": "https://arxiv.org/pdf/2206.01062.pdf"}'
```

The duplicate `libavdevice` dylibs warning is harmless noise from Gradio/CV2 conflicts, but the EasyOCR error is fatal.
```

#AIHype #TheAICON #Anthropic

Build Agent-Ready RAG Systems in Java with Quarkus and Docling https://www.the-main-thread.com/p/enterprise-rag-quarkus-docling-pgvector-tutorial

#java #docling

Barrierefrei für Mensch und Maschine – Kleine Helferlein zur Konvertierung von Office-Dokumenten in maschinenlesbare Formate

Die „Währung“ in der dokumentiertes Wissen in Organisationen transferiert wird sind Office-Dateiformate wie pptx, docx, xlsx und PDF. Leider sind das Formate, die für Maschinenlesbarkeit denkbar ungünstig sind. Oft sind sie zusätzlich so groß (z.B. durch in Powerpoint-Präsentationen eingefügte Grafiken), dass sie schnell die Limits von KI-Chatbots wie Copilot, ChatGPT und Claude […]

#docling #genai

https://cogneon.de/?p=66598

Tomorrow (Feb 4) at #CfgMgmtCamp in Ghent, Ming and I will run a workshop on #Docling at 14:00 in B.1.031 - https://cfp.cfgmgmtcamp.org/ghent2026/talk/9CV7CY/

Join us! @cfgmgmtcamp

The slides and recording for my presentation on Get your docs in a row with #Docling are now available - https://fosdem.org/2026/schedule/event/DVRV8S-get_your_docs_in_a_row_with_docling/

Thanks to @fosdem organizers and volunteers for another amazing event. My 11th in-person #FOSDEM (13 including virtual ones).

The docling-testcontainers module provides a ready-to-use Testcontainers integration for running a Docling Serve instance, wrapping the official container image and exposing a simple Java API.
https://testcontainers.com/modules/docling/

#Docling #Java #Testcontainers

Как затащить AI в Java/Kotlin проект

Мир Enterprise-разработки на Java/Kotlin и мир нейронных сетей кажутся параллельными вселенными. С одной стороны - статическая типизация, многопоточность, Spring-контейнеры, а с другой - Python-скрипты, тензорные операции и эксперименты в Jupyter Notebook. Между ними - пропасть, через которую многие команды не решаются перешагнуть. Однако необходимость строить этот мост возникает всё чаще. Заказчик хочет «искусственный интеллект» в новом фиче, аналитики мечтают о реализации чат-бота с преферансом и барышнями, а менеджеры слышали, что конкуренты уже всё автоматизировали. Как же совместить надежность и структуру JVM-проекта с гибкостью и мощью AI? В этой статье постараемся разобраться какие инструменты для этого есть на данный момент и как с ними работать.

https://habr.com/ru/articles/984544/

#AI #ИИ #Java #Kotlin #LLM #State_Graph #Vector_DB #Docling #Embeddings

Just published a new deep-dive on building enterprise-grade RAG in Java.

In this tutorial, we combine:

• Quarkus
• Docling (layout-aware PDF parsing)
• pgvector + PostgreSQL
• Local LLMs via Ollama
• And a simple guardrail layer

This is the most complete RAG pipeline I’ve built so far, and it’s fully open for you to copy, run, and adapt.

Read here:
https://www.the-main-thread.com/p/enterprise-rag-quarkus-docling-pgvector-tutorial

#Java #Quarkus #LLM #RAG #Docling #OpenSource