Content for workshops on computer vision @ HPI's AI Service Center

Harnessing Large Language Models for Curated Code Reviews

Implementation of TSDS: Data Selection for Task-Specific Model Finetuning. An optimal-transport framework for selecting domain-specific and task-specific training data to improve LLM finetuning and instruction tuning.

HyperView curates datasets and provides model introspection in hyperbolic and Euclidean geometries.

[ACL 2024 (Findings)] ICC: Quantifying Image Caption Concreteness for Multimodal Dataset Curation

Image description/tagging tool

NAACL 2025 | How Can We Diagnose and Treat Bias in Large Language Models for Clinical Decision-Making?

An image deduplication GUI, made for image generation models dataset deduplication using CLIP.

Manage and process paired RGB and depth images, with options to view, export, and exclude images using various colormaps.

Comprehensive framework for curating and validating biomedical datasets for clinical AI applications

Biomedical Image Processing BAP (Scientific Research Project) - Piri Reis University

AIWG training-complete framework — corpus-to-dataset pipeline with SKILL.md agentic surface and optional Python runtime backend. Marketplace plugin for AIWG.

Pipeline for querying and turning NASA's ADS publications metadata into curated, analysis-ready datasets, topic maps, and citation networks.

A local-first self-improvement runtime for language systems. Record. Learn. Rewrite.

Workflow and validation toolkit for human review of wildlife AI outputs

ML/AI Data Curation Functional Setup

Two-stage video captioning pipeline: a Vision-Language model produces a rich description, then a text-only LM rewrites it through a task-specific prompt (e.g. for LoRA training datasets, retrieval, summarization).

Efficient, reproducible dataset curation for LLM fine-tuning: scripts and best practices for preparing code datasets without repository bloat.

🪘 Tabla Drum Image Generator – AI-powered tabla drum image generation using Stable Diffusion & GANs. Features custom dataset curation, ML training pipeline, and scalable API deployment.

GPT-2 fine-tuned model for myth generation using curated mythological datasets and structured NLP preprocessing.