What's on My HomelabA self-hosted infrastructure stack running 40+ services on bare-metal Kubernetes — ERP, project management, AI, messaging, and more — for a few hundred dollars a month instead of $20,000+ on AWS.
Data Gateways: Turning Third-Party APIs into a System of RecordWe build small services that poll third-party REST APIs and publish everything to Apache Pulsar — turning external APIs into a replayable, event-sourced system of record we fully control. Here is the
Nix: Too Hard for Humans, Perfect for AI AgentsNix is too hard for humans to use at scale — but its properties of hermeticity, reproducibility, and declarative composition make it an ideal substrate for AI coding agents.
Building a Self-Dispatching Coding Agent FleetAutonomous coding agents that pick work from a project board, execute it, and report back — no human in the dispatch loop. Pull-based coordination, optimistic locking, and Epic ownership explained.
Scaling MonosemanticityThe Scaling Monosemanticity paper explores sparse dictionary learning to extract interpretable features from large language models, with applications for AI safety and model steering.
Modeling Evolution in Probabilistic Ontology-Driven Multi-Agent Systems (pODMAS) with Gaussian Mixture Models (GMMs)Representing entity evolution through bitemporal Gaussian Mixture Models — tracking both real-world changes and system knowledge updates in multi-agent systems using adaptive probabilistic modeling.
Using Reinforcement Learning to Guide the Thought Process of Large Language Models Over Multiple Computation StepsExploring how reinforcement learning can guide LLMs through multi-step reasoning tasks — making models more reflective, goal-directed, and efficient in complex problem-solving and dialogue systems.
