I build production LLM systems in the open — multi-model orchestration, MCP servers, evals in CI — and I've spent two decades running and transforming technology operations in global banking, currently as Director, SRE and Operations Transformation. The open-source work is where the two meet: applying reliability engineering discipline to AI systems.
- llm-council — multi-LLM deliberation with anonymized peer review and synthesis. Exists because single-model answers fail silently; a council disagrees out loud. Python library, MCP server, or HTTP API — llm-council.dev.
- llm-council-action — the council as a GitHub Action quality gate, because LLM judgment belongs in CI, not just in chat.
- luminescent-cluster — three-tiered memory architecture for persistent AI coding context. Exists because agents forget, and context is the expensive part.
- amiable-templates — production-ready Railway deployment templates for AI infrastructure, so the ops half of "LLM ops" isn't an afterthought.
- conductor — multi-protocol input automation for MIDI and custom hardware (Rust + WASM plugins).
Currently in the workshop: an evidence-grounded SRE assessment engine — deterministic policy gates, advisory LLM, gold-dataset evals in CI. Open sourcing when the core is finished, not before.
Two tracks at amiable.dev: reliability engineering in regulated environments (SLOs, incident engineering, operational resilience under DORA/FCA) and LLM-ops (evals, MCP, agent reliability).