AI observability as the engine of production-ready AI

Samuel Jaja, Generative AI Engineer at Capgemini UK, gives a talk titled "From Black Box to Glass Box: Systems Thinking & Observability as the Engine of Production-Ready AI" at Leeds Sharp's May meetup. The talk uses a real multi-agent financial compliance system as its central case study: five specialised agents handling OFAC sanctions screening, AML detection and regulatory rules enforcement, with every agent decision captured in an immutable Langfuse audit trail showing token usage, latency, cost and data lineage in real time.

Jaja walks through three architecturally distinct ways to build the same system. First, raw SDK with custom parallel orchestration using asyncio. Second, the OpenAI Agents SDK for rapid prototyping. Third, MCP (Model Context Protocol) for exposing tools across team and organisational boundaries. Each approach trades differently, and the talk is explicit about when to reach for which. The session also covers Azure OpenAI's role when data sovereignty and auditability are non-negotiable in regulated industries such as finance, energy and construction.

Attendees will leave with a concrete mental model for when multi-agent architecture is genuinely necessary rather than fashionable, and why observability functions as the feedback loop that compresses build-debug-ship cycles rather than something bolted on afterwards.

Leeds Sharp is a long-running .NET user group based in Leeds. It meets monthly at venues around the city centre, drawing engineers who work across the .NET ecosystem. Sponsors include Fruition Group, JetBrains and tronik. The group is hosted by JohnS and has a track record of bringing in speakers with genuine production experience rather than conference-circuit abstractions.

This is the group's first event at Enablis on York Place, so note the new venue before you set off.

Why attend

Jaja has built production AI systems in finance, energy and construction at Capgemini UK, and founded PyData Hull, the youngest NumFOCUS chapter in Yorkshire. He spoke at DDD North 2026. This is not a theoretical overview of observability tooling — it is a walkthrough of a real compliance pipeline with real architectural trade-offs between three distinct SDK approaches.

The evening starts at 6pm with pizza and drinks before the talk begins at 7pm. Some attendees head to the Editor's Draft pub afterwards. If you work in .NET, AI engineering or regulated-industry software, this is a practical evening with people doing similar work in Leeds.

Agenda

• 18:00 — Doors open, pizza and drinks
• 19:00 — Samuel Jaja: "From Black Box to Glass Box: Systems Thinking & Observability as the Engine of Production-Ready AI"
• 20:00 — Close; optional drinks at the Editor's Draft pub (vote taken on the night)

Speakers

• Samuel Jaja — Generative AI Engineer at Capgemini UK; founder of PyData Hull and Atlasync AI Ltd; MSc in AI and Data Science with Distinction from the University of Hull; speaker at DDD North 2026.