I build systems where machine learning doesn't just analyze the world - it participates in it. From microrobot swarms in flowing blood to AI chemists parsing molecules, my work fuses artificial intelligence with the natural and exact sciences to push discovery further, faster, and smarter.
I'm Josef Berman, a data scientist and researcher building bridges between artificial intelligence and the natural world. My work lives at the intersection of multi-agent systems, scientific discovery, and intelligent automation.
At the heart of everything I do is a simple idea: AI shouldn't just analyze science - it should participate in it. I create agentic systems that interact, adapt, and collaborate - whether they're learning to navigate complex physical environments, forecasting air pollution, or decoding the structure of molecules. Think swarms of autonomous microrobots, AI chemists parsing spectroscopic data, or neural networks predicting environmental shifts.
With a background spanning reinforcement learning, computational physics, and data engineering, I design systems that are grounded in the laws of nature but powered by the mathematics of intelligence.
Technologies: Python, PyTorch, TensorFlow, scikit-learn, LangGraph, LangChain, PhiFlow, RDKit, Stable-Baselines3, Gradio
Reinforcement learning meets fluid dynamics: FluxSwarm trains microrobot swarms to maneuver through dynamic environments like blood vessels or pipelines. The system learns to resist flow, collaborate under pressure, and adapt to nonlinear conditions - bridging bio-inspired robotics and AI.
Using gradient boosting and cheminformatics, this tool infers molecular substructures directly from 1D NMR spectral data. It aims to automate and accelerate the process of structure elucidation, offering interpretable and efficient insights for chemists.
syN-BEATS is a neural ensemble model designed to predict airborne pollutant concentrations with limited sensor data. Built on the N-BEATS architecture, it achieves strong performance even in under-monitored regions - empowering environmental agencies with sharper forecasting.
A retrieval-augmented generation (RAG) system fused with molecular property predictors to uncover hidden features of explosive compounds. EnergeticGraph links semantic search, SMILES translation, and graph-based inference to support chemical safety and synthesis research.
Exploring how autonomous AI agents could revolutionize the way we conduct scientific research in the next decade. From automated hypothesis generation to robotic lab assistants.
Read PostI'm always open to discussing new projects, research collaborations, or opportunities to apply data science to interesting problems.
Send Email