Robert Haase
LightNews LightNews
banner
haesleinhuepf.bsky.social
Robert Haase
@haesleinhuepf.bsky.social
4.1K followers 1.2K following 280 posts
Biomedical Computer Vision & Language Models, scads.ai / Leipzig University, also NFDI4BIOIMAGE, NEUBIAS/GloBIAS, GPUs, AI, ML 🔬🖥️🚀 views:mine
Posts Replies Media Videos
haesleinhuepf.bsky.social
I just finished my @unileipzig.bsky.social #LunchLecture. Thanks everyone for the vivid discussion and Nicole Koburger for hosting us!
The slides are online #openaccess 🧑‍🏫🔓 zenodo.org/records/1742...
October 23, 2025 at 11:19 AM
haesleinhuepf.bsky.social
Fantastic question! Here you go: github.com/haesleinhuep...
October 7, 2025 at 10:53 AM
haesleinhuepf.bsky.social
For academic purposes, I generated 100 UMAPs from the same tabular raw data, I sorted them and the result looks too funny to not share it with you 😃
haesleinhuepf.github.io/BioImageAnal...
October 1, 2025 at 2:17 PM
haesleinhuepf.bsky.social
I gave two seminars at @embl.org Heidelberg these days: One about using LLMs via #biabob for #BioImageAnalysis and one about how LLMs impact BioImage #DataScience
Thanks EMBL for having me!
Materials #openaccess 🧑‍🏫🖥️🔓
zenodo.org/records/1705...
zenodo.org/records/1710...
scads.github.io/llm-bia-2025
September 16, 2025 at 7:02 PM
haesleinhuepf.bsky.social
I'm sitting in a train with internet connection quality like everyone would expect it in central Germany, and I would like to thank ollama.com for its existence! 🤩
This response (< 5min, video in real-time) is generated using #ollama running locally on my laptop with 32GB Ram and an RTX 4070 GPU 🤯
September 11, 2025 at 11:42 AM
haesleinhuepf.bsky.social
September 5, 2025 at 8:19 PM
haesleinhuepf.bsky.social
🫠
A plot with "time point" on the x-axis and "normalized envelope intensity" on the y-axis, and a single data point at 0 (time) / 1 (intensity).
September 3, 2025 at 1:42 PM
haesleinhuepf.bsky.social
Check out this masterpiece of systematic #BioImageAnalysis 🤩
August 30, 2025 at 9:34 AM
haesleinhuepf.bsky.social
Just preparing a hand-out for my next talk😃
August 28, 2025 at 7:33 AM
haesleinhuepf.bsky.social
AI-assisted cherry picking 🍒 using
next gen #biabob 🚀 #sneakpreview
August 25, 2025 at 1:40 PM
haesleinhuepf.bsky.social
My tutorial materials & slides about #Reproducible #BioImageAnalysis using Python, Napari, Jupyter and #AI are available #openaccess.
Thanks to #EMBOlsm25 for having me! It so great to visit friends at my former academic home campus for a couple of days 🤩
scads.github.io/embo-lsm-bia...
August 12, 2025 at 5:57 PM
haesleinhuepf.bsky.social
Curious: Why do you think are these two outputs soo similar?
August 10, 2025 at 2:02 PM
haesleinhuepf.bsky.social
SVG is a good choice. Result quality depends on the model obviously. github.com/haesleinhuep...

Also check out @simonwillison.net's blogs posts about LLMs creating SVGs, some are hilarious simonwillison.net/tags/svg/
August 9, 2025 at 12:44 PM
haesleinhuepf.bsky.social
Some folks wonder how many B's are in blueberry. 🤷
#biabob
August 8, 2025 at 5:11 PM
haesleinhuepf.bsky.social
After we lost our beloved cat Pixel earlier this summer, we decided to increase our catabilities:
Please welcome Oskar (left) and Kiwi (right) #kittens 🐈🐈‍⬛️
July 27, 2025 at 7:43 AM
haesleinhuepf.bsky.social
I asked ChatGPT to answer my exam questions and compare its responses to mine. Now I’m just trying to figure out who passed the test — ChatGPT or me. 🫠
July 14, 2025 at 12:05 PM
haesleinhuepf.bsky.social
The @globias.bsky.social #preprint 🚨 is out, outlining global needs of BioImage Analysts, presenting how GloBIAS seeks to support and professionalize this role through education, and community building! Congrats Agustin Corbat and the entire team for pulling this through!
arxiv.org/abs/2507.06407
July 10, 2025 at 1:19 PM
haesleinhuepf.bsky.social
New in #biabob 0.29.0: Code insertions in @jupyter.org Cells! 🥳
If you know how to start your image analysis, and how to visualize the results, just put %bob in between and let the AI do the hard coding work. Also provide hints as comments to guide the AI 😉
github.com/haesleinhuep...
July 3, 2025 at 12:21 PM
haesleinhuepf.bsky.social
I just finished a full-day hands-on tutorial about applications of LLMs as part of the @scadsai.bsky.social Summer School. Big thanks everyone for the lively discussions about prompt engineering, code&data generation, RAG, vision, function calling and #AI agents 🖥🤖🚀
scads.github.io/LLM_Summer_S...
ScaDS.AI Summer School LLM Training This is a collection of Jupyter Notebooks about basic and advanced LLM usage, focusing on what’s practicslly possible with Large Language Models. We demonstrate most utilities using the ScaDS.AI Large Language Models (LLM) Servicer (TU Dresden VPN required). Most exercises should also work with other providers such as GWDG KISSKI and Helmholtz Blablador. The tutorial aims at Python programmers who want to dive into LLMs for generating text, code and data using open source/weights models. After this course they will be able to integrate LLMs into their software applications. Contributions and feedback are very welcome! In case you see room for improvement, please create a github issue and/or consider contributing. Topics The notebook collection aims covering these topics: Large Language Models (LLMs) Text/Code/Data generation Prompt Engineering Retrieval-augmented-generation Vision Function calling Agents Covered Python libraries and software In these notebooks we use libraries from the GenAI field: bia-bob llama-index ollama openai Slide decks Also training slides can be downloaded from zenodo: https://doi.org/10.5281/zenodo.15743126 Videos The materials provided here were also discussed in a couple of recorded Online Lectures Large Language Models for Bio-image Analysis (ScaDS.AI Living Lab lecture 2024) Large Language Models – an introduction for life scientists (GloBIAS Seminar 2024) Bio-image Analysis Code Generation (I2K Conference 2024) Generative Artificial Intelligence for Bio-image Analysis Related resources The notebooks presented here were partially taken from other places or inspired by code in these resources:
June 26, 2025 at 5:03 PM
haesleinhuepf.bsky.social
It was an honor to give a keynote lecture about #VLMs for Bio-Image #DataScience at the @helmholtzimaging.bsky.social #HIconference2025.
You find my slides #openaccess 🔬🖥️🚀
doi.org/10.5281/zeno...
June 25, 2025 at 5:19 PM
haesleinhuepf.bsky.social
At the AI MeetUp (KI Stammtisch) today at Universitätsbibliothek Leipzig I gave a short intro to the #EUAIAct. My slides are available #openaccess in English and German 🧑‍🏫 🖥️ 🔓
doi.org/10.5281/zeno...
June 19, 2025 at 2:53 PM
haesleinhuepf.bsky.social
If you seek for an AI coding buddy to review your code and provide umprompted feedback, try out _unprompted_! I hope it helps to identify issues in code early! 🖥️🤖💫
pip install unprompted
github.com/haesleinhuep...
June 15, 2025 at 9:16 AM
haesleinhuepf.bsky.social
Are you curious how privacy-preserving LLM-based AI-assistants work that can answer emails?
Check out outlook-bob! It's uses local privacy-preserving LLMs, it's #opensource, surprisingly simple 😀
github.com/haesleinhuep...
June 12, 2025 at 7:18 AM
haesleinhuepf.bsky.social
#biabob 0.28.0 is out and it comes with support for OpenAI's codex-mini model. I also enabled its vision-support. With this, you can turn screenshots of user-interfaces into functional code - directly within your Jupyter notebook 🤯
github.com/haesleinhuep...
June 1, 2025 at 10:57 AM
haesleinhuepf.bsky.social
If you use ML for pixel classification, e.g. using sckit-learn, you can use SHAP analysis for explaining how relevant the feature images are you provided.
I just added a Notebook demonstrating how to do this to the #BioImageAnalysisNotebooks 🔬🖥️📈
haesleinhuepf.github.io/BioImageAnal...
May 20, 2025 at 7:35 AM