- 🛡️ Deterministic Agents: Human in the loop control, fixed execution paths, and audit trails for autonomous systems
- 🛩️ Commercial Aerospace: Orbital mechanics, electric propulsion, and space instrument modeling in MATLAB, Python, and C++
- 🌀 World Models: Evaluation harnesses for latent geometry in action conditioned video world models and in clinical state spaces
- 🏥 Multimodal Medical LLMs: Healthcare AI and diagnostic systems
- 📐 Hyperbolic Geometry for AI: Latent spaces whose geometry matches the hierarchical structure of the data they encode
- 31+ datasets across photonics, healthcare, aerospace, and legal domains
- Ongoing work on aerospace and photonics datasets and hyperbolic representation learning, with world models as the first evaluation domain
- Publishes models and datasets on Hugging Face
- Contributor to the Hugging Face Computer Vision Course
- 🚀 Aerospace AI: Mission profile sims, electric propulsion modeling, with MATLAB and C++ implementations alongside Python
- 🏥 Healthcare AI: Medical imaging and diagnostic datasets, and geometry aware evaluation of clinical world models
- 🤖 Multimodal Gen AI: Cross modal learning and synthesis
- 💎 Integrated Photonics: SiN waveguide datasets and thermal drift simulation for deep space based instruments
- 📐 Latent Geometry: Hyperbolic and Riemannian latent spaces for world models
- 🌀 Hyperbolic-World-Model: Retrains the action conditioned predictor head of a frozen video world model (V-JEPA 2-AC, with DINO-WM as a second subject) in Euclidean, Poincaré, and Lorentz latent spaces, and measures whether negative curvature improves rollout error, hierarchy recovery, and dimension efficiency.
- 🏥 Clinical-Geodesic: Evaluation harness for medical world models, applying the same frozen encoder and native geometry methodology to clinical state prediction.
- 🎯 Application targets: Clinical trajectory prediction, and fuel and time efficient space probe transfers to the asteroid belt and Jovian planets.
- 🛡️ Agent Harnesses: Deterministic execution, dry run safety gates, and tamper evident audit logs for production agents
- 🤖 Agentic AI: Production multi agent systems on Watsonx Orchestrate and Amazon Bedrock; evaluating Gemini Enterprise Agent Platform
- 🛠️ Tech Stack: Anthropic Claude (API and Agent SDK), AWS Strands Agents, Google ADK, MCP
- 🌀 World Models: Geometry aware evaluation harnesses (PyTorch, geoopt, Hydra) with CI and reproducible one command reports
- 📐 Hyperbolic Latent Spaces: First evaluation of the hyperbolic representation thesis, comparing action conditioned world models
- 🔌 MCP: Building MCP servers and tooling; contributor to IBM Context Forge
- 🚀 Scientific Computing: Physics libraries (MATLAB, Python, C++20) with cross language numerical parity and benchmark validation
- 🛡️ Deterministic agent harnesses: Self verifying workflows and human approval gates for production deployments
- 🌀 World models: Latent geometry evaluation, hyperbolic embeddings, and benchmark design for embodied and clinical state prediction
- 🤖 Multimodal autonomous agents: Portable agent designs across Bedrock, Watsonx Orchestrate, and ADK
- 📐 Hyperbolic representation learning: Hyperbolic latents for hierarchical and sequential structure, starting with world models
- 💬 Commercial Space Industry discussions and Propulsion insights
- 🗣️ German Language Practice: Seeking relaxed conversations with native speakers in 2026
