Explore projects

This repository accompanies a three-day hands-on workshop on Large Language Models (LLMs), Prompt Engineering, and Retrieval-Augmented Generation (RAG). The workshop is designed to equip participants with practical skills and foundational knowledge to understand, deploy, and evaluate LLM-based applications across a wide range of fields. The program combines a lecture series covering theoretical foundations with a programming workshop in Jupyter Notebooks. Participants explore core technologies such as OpenAI APIs, LangChain, vector databases, document parsing, and AI toolchains—supported by real-world examples and scientific documents. The workshop is suitable for professionals, researchers, and students who seek a structured and practice-oriented introduction to state-of-the-art AI workflows based on LLMs.

The Prototype Document Extractor is a lightweight, containerized service designed to extract structured content from PDF files using the Unstructured IO library. It exposes a minimal HTTP API that allows users to submit PDFs and receive parsed content in JSON format. This project includes:

A backend service that handles PDF parsing using Unstructured IO. A Python client library for programmatically interacting with the API from within your code. Docker configurations to run the service in a portable, reproducible environment.

This project provides a Dockerized setup for running Weaviate as a local vector database. It includes a pre-configured Docker Compose setup, support for manual vector handling, and optional integration with vectorization modules using OpenAI or a local embedding model with Ollama. The setup features an example Jupyter notebook demonstrating Weaviate’s CRUD workflow and client library usage. Configuration is managed via .env, ensuring flexibility for different vectorization approaches. Designed for quick deployment, this setup serves as a practical starting point for building AI-powered applications with vector search capabilities.

automatic processing of objections; Keywords: uba, umweltbundesamt, ki-lab, NLP, LLM, FKTG, Endlagersuche

automatic evaluation of an ITZBund Chatbot using a large language model; Keywords: uba, umweltbundesamt, ki-lab, Bundesbot, LLM

Convert Greta output data to match the CLRTAP gridded data template

Constraining national emission inventories using earth observation data

Proaktiv handeln, statt nur reagieren: Das in LeAn® integrierte Frühwarnsystem ist das erste Analytics Tool für Stadtentwickelnde, die ihre Stadt von der datengetriebenen Seite kennenlernen und verstehen wollen. Auf Basis relevanter Kennzahlen werden drohende Leerstände in einem ganzheitlichen Kontext betrachtet, bewertet und interpretiert, um ihnen langfristig präventiv begegnen zu können.

Using ePRTR data for CLRTAP large point source reporting