Physical AI Midlayer

private

Intent-based robotics R&D platform — multi-agent orchestration, embedded AI on NVIDIA Jetson, ROS2 ⇄ MCP interop.

2025-10 – Present · @ Device Insight GmbH

Python
LangGraph
Microsoft Agent Framework
MCP
vLLM
NVIDIA Jetson
TensorRT
quantization
ROS2

An R&D platform built for the CTO Office of a leading robotics manufacturer, sitting between high-level customer task intent and low-level robot behaviour.

What I'm doing

Solo engineer translating ambiguous CTO-Office briefs into working prototypes, architecture proposals, and platform-evolution decisions. Built a working multi-agent orchestration prototype running on NVIDIA Jetson — evaluated LangGraph and Microsoft Agent Framework hands-on, prototyped on-device inference (vLLM, quantization) and error-tolerant task execution. Currently bridging the robotics-engineering side of the platform with the agentic orchestration layer via ROS2 ⇄ MCP interop.

Authored an 80-page internal whitepaper that drove platform- evolution decisions — selected which components advance to next-stage research, which architectural directions the platform takes.

Source is private. This entry exists for context.