AI platform engineer building multi-domain intelligence systems.

I design and build the Opsight Intelligence Platform — a federated system where independent domain engines (fraud, maritime, manufacturing, AI-governance) each run their own collect-to-assess pipeline and publish a single standardized intelligence contract to a shared bus, which a federation gateway exposes to LLM-driven executive AI personas. The focus is contract-first architecture: one shared schema, strict pure-core / thin-I/O layering, and engines that stay independent yet speak the same language.

I build AI systems the way I build backends — with real engineering discipline: grounded, governed, testable, and reliable in production.

• AI systems — LLM-driven agents, MCP servers, RAG pipelines, and executive AI personas that consume structured assessments; prompt design, structured outputs, tool/function calling, and AI governance (grounding, human-in-control, evaluation).
• Platform architecture — contract-first design, dependency inversion, pure business logic isolated from I/O adapters, single-source-of-truth registries.
• Intelligence pipelines — collect, normalize, enrich, detect, score, assess, publish; idempotent, content-addressed artifacts.
• Distributed services — Python backends, an event/intelligence bus, MCP servers and a federation gateway.
• Engineering discipline — small modules, full type hints, real CI gates (lint + types + tests), reproducible releases.

Aligning every Opsight domain engine onto the Intelligence Bus, building out the MCP federation gateway, and moving the executive AI personas to consume structured assessments instead of raw documents.

Active development happens in the Opsight Intelligence organization, where the domain engines, intelligence bus, and federation gateway are built. Most of those repositories are private, so for a public, high-level view of the architecture and design principles, see:

→ Opsight Architecture Overview

This profile focuses on the platform-level thinking; the overview repo describes the system without exposing implementation details.