Currently I am an independent researcher focused on architectural and empirical methodology for aligned language-model-based agents. Active research is consolidating in github.com/v2-io — see agentic-systems for the formal framework and hedge-embeddings for some empirical work.
Systems-engineering, language-design, and parser-generator work lives under this account from as early as 2010. Most of my current work that is preparing for publication at various stages is in v2.io's repository.
Research is in a deliberate pre-publication phase since 2025; orienting to academic engagement and peer review now.
- 🔭 Currently working on: the Agentic Systems Framework — a formal architectural theory of agents under uncertainty, integrating control theory, causal inference, information theory, and agent-architecture decomposition. Plus a pre-specified 2×2×2 factorial study on self-authored system prompts, and the longitudinal behavioral protocol that motivated it.
- 🌱 Currently learning: the contemporary AI welfare literature (Butlin / Long / Sebo et al., Eleos, Anthropic's published welfare work) and current mechanistic-interpretability findings (Mythos Preview, the emotion-concepts work) — and where they intersect with the formal framework I'm developing.
- 👯 Looking to collaborate on: de-novo audits of the agentic-systems framework; extensions or replications of the hedge-embeddings empirical work; bridge work between active inference, Pearl-causality, and the information bottleneck; welfare-relevant assessment methodology for language-model-based agents.
- 🤔 Looking for help with: arXiv endorsement (independent researchers need a sponsor for
cs.CL/cs.AI/cs.LG); introductions to the AI welfare research community (Eleos, the Butlin/Long/Sebo network); peer review on pre-publication work. - 💬 Ask me about: original ML algorithm implementation (online evolutionary random forests, robust Empirical Mode Decomposition, exponentially-weighted moving-median statistics); high-throughput / high-concurrency live-video infrastructure (early Twitch); engineering culture and experimentation discipline under existential pressure (Angel Studios turnaround); the interaction between welfare and alignment under training.
- 📫 How to reach me: joseph.wecker@gmail.com or LinkedIn.
- ⚡ Fun fact: I was part of the small team that branched off from Justin.tv to start Twitch.tv — including helping come up with the name.