About

I'm Chris Joseph — Director, SRE and Operations Transformation at a global bank, and an AI systems engineer in the open. I've spent two decades building and operating technology where reliability is a regulatory and commercial obligation, and I now apply that discipline to production LLM systems: multi-model orchestration, MCP servers, evals in CI. This site is where the two meet.

What I focus on​

• Reliability in regulated finance — SLOs and error budgets for regulatory processes, incident engineering, and the operational-resilience practices that DORA and the FCA/PRA regimes now mandate; controls that auditors and engineers can both live with.
• LLM-ops & agent engineering — evaluation frameworks, MCP servers, multi-model deliberation, and agent reliability, built as open source.
• Observability & platform engineering — monitoring that reflects business outcomes, OpenTelemetry in practice, alert quality, and delivery platforms built with production-grade rigour.

Career​

I started as an analyst programmer in the late nineties — AS/400 systems, MQSeries integration, Y2K remediation — then spent the 2000s in consultancy and financial-services delivery, from core-banking interfaces to a quotation-management platform for the reinsurance market.

From the mid-2000s I moved into banking technology proper: wealth-management platforms, then a decade leading Finance IT application development for a global capital-markets business — regulatory reporting, liquidity risk, and Basel III programmes where a missed deadline is a reportable event.

That regulatory-delivery background pulled me toward the operational side: building an IT risk management function across 100+ applications, then establishing DevOps enablement that cut deployment times by up to 90% across 36 applications, and an operations event bus — Kafka, AIOps enrichment, automated routing and escalation — that reduced mean-time-to-everything for production incidents.

Today I lead SRE and operations transformation: SRE squads and guilds, SLIs, SLOs and error budgets applied to regulatory processes (a 40% reduction in process instability), and continuous-improvement programmes grounded in production data. Outside the day job, I build open-source LLM infrastructure — deliberation systems, evaluation gates, agent memory — because the reliability problems in AI systems are the same ones, wearing new clothes.

How I work​

I believe engineering decisions should be visible and reviewable. This site applies that principle to itself:

• Production monitoring, in public — the status page is driven by the same incident-automation patterns I write about: health checks every five minutes, incidents filed and closed as GitHub issues, uptime history published.
• Transparent AI assistance — some posts here are written with AI assistance. Every one of them says so explicitly, with the model, the review process, and quality metrics in the post metadata. I treat provenance the way I treat observability: if you can't see it, you can't trust it.