Stochastic Agents
LiveStochastic Agents is a research collaboration between SAP and TUM that challenges the standard orchestrator-agent pattern. Instead of a single controller directing agents top-down, the system explores how networks of agents can self-organise, negotiate, and collaborate to solve tasks — closer to how distributed teams work than how pipelines do. I owned the DevOps and infrastructure end-to-end: containerisation, CI/CD pipelines, and deployment on Railway, while contributing to the core agent communication layer.
1 / 2
Features
- →Agent-to-agent communication without a central orchestrator
- →Dynamic task delegation across a network of specialised agents
- →Containerised deployment pipeline on Railway
Challenges
- →Keeping agent coordination stable under stochastic, emergent behaviour
- →Building a reproducible deployment pipeline for a fast-moving research codebase