I build production AI systems — agentic platforms, coding-agent harnesses, LLM evaluation and guardrails, and the platform engineering that keeps them running. I'm Principal Architect at Presidio, with 25 years across engineering and architecture, including six years as a Solutions Architect at AWS and a turn as CTO of a streaming platform.
I build production AI systems that engineering and revenue teams actually use — not demos, not strategy decks.
Selected work
Sanitized for public publication. Customer names, internal product names, employee names, and specific deal amounts removed. Patterns and methods preserved.
AI engineering platform (in production, 9+ months)
An AI-native operating system for enterprise presales pursuits — covering discovery, qualification, intel, proposals, SOW redlines, risk review, meeting prep, and engagement handoff on a single context store.
Scope: 16 plugins spanning the full pursuit lifecycle, exposing 95 slash
commands, 22 skills, and 6 subagents behind one operator surface. Each capability ships
as an independent plugin; a shared core provides the persona, path resolver,
and rules everyone else depends on.
Engineered like a platform, not prompt glue:
- Prompt-contract framework with versioned input/output schemas — contract violations surface in the developer console instead of degrading silently
- Evidence-citation guardrails: every claim carries a High/Medium/Low confidence rating and a dated source URL, or the output refuses to generate
- A comprehensive automated test suite plus plugin-validation and release-automation tooling — the marketplace is shipped, versioned, and verified like any other software product
- Multi-provider model routing (Anthropic Claude primary, AWS Bedrock fallback), switchable by environment variable
- Prompt caching for shared guardrails and the knowledge base, with a token-budget discipline that keeps the corpus lean as it grows
- Session-start hooks hot-load operational knowledge from synced channels — institutional memory becomes ambient without a release
Outcomes (anonymized):
- Compressed manual artifact production hours into command-driven workflows
- Enabled $0.5M+ and $1M+ enterprise engagements to close on platform-authored statements of work with iterative revisions
- Partner-funded engagement program added co-funding to a pursuit
- Productized AI Well-Architected workshop at small-five-figure list price; first paying customer in flight
The pattern: start shipping custom for one customer, then extract patterns into a framework for ten field engineers.
Customer-facing pursuit microsite pattern
Built a customer-facing pursuit microsite in days for an enterprise AI agent engagement in K-12 EdTech using Claude-on-platform. The microsite anchored follow-up conversations with the client's CEO and hyperscaler co-sell partner; the pursuit moved from ideation workshop to verbal Phase-1 commitment in three weeks.
Pattern observations:
- AI-listener-agent architecture on AWS Bedrock + ECS Fargate behind WebSocket API Gateway, with OpenSearch Serverless for RAG and Transcribe/Polly for voice
- POC built live in ideation workshop (single-file Flask + Anthropic, ~3,700 lines, eight views) served as conversation piece pulling stakeholders into architecture discussions faster than slides
- Microsite + POC + workshop collapsed what used to be three serial deliverables into parallel motion
The artifact pattern is portable: AI pursuit microsites are cheap to build, land differently than decks, and serve as persistent reference points through Phase-1/Phase-2 sequencing.
Token-efficient Claude Code corpus design
A reproducible methodology for keeping a multi-plugin Claude Code corpus lean. Took a production corpus from 221K tokens → 179K tokens (19% reduction) without removing procedural steps or contract terms.
| Pool | Before | After | Saved | % |
|---|---|---|---|---|
| Commands (per invocation) | 146,105 | 111,140 | −34,965 | −23.9% |
| Rules (auto-load every session) | 18,375 | 14,478 | −3,897 | −21.2% |
| Skills (per invocation) | 56,620 | 53,815 | −2,805 | −5.0% |
| Total | 221,100 | 179,433 | −41,667 | −18.8% |
Seven patterns that drove the savings: rules dedup (byte-identical context files collapsed into a shared core/rules/); trimming the right-skewed top decile of commands (42% of the corpus); removing repeated pipeline-banner step counters; collapsing auto-resolve fallback boilerplate to one sentence; one base template + deltas for multi-format outputs; dropping redundant persona-reload reminders; and leaving load-bearing skill specs untouched. Token counts via tiktoken / cl100k_base.
"Lean prompts compound. Cluttered prompts compound differently."
Selected AI engineering primitives
- Cross-platform OneDrive path resolver — disambiguates personal vs. organizational mount paths across macOS and Windows; survives the OneDrive rename/move dance that breaks most tooling.
- Prompt-contract framework — versioned input/output schemas per command, with contract violations surfacing in the developer console rather than degrading silently.
- Evidence-citation + confidence-disclosure guardrails — every claim must cite a dated source URL and a High/Medium/Low confidence rating; outputs refuse to generate if they can't satisfy the contract.
- Session-start hook that hot-loads operational knowledge — institutional memory becomes ambient without requiring a plugin release.
- Multi-provider AI infrastructure — Anthropic Claude (primary) with AWS Bedrock fallback; prompt caching; per-command rule contracts; provider switching via env-var toggle.
:councilmulti-agent advisory pattern — pressure-tests recommendations through a 5-advisor council before reaching the user; surfaces dissent that single-agent flows hide.- Document generation pipeline — branded Microsoft Word output via AI-augmented markdown rendering; preserves enterprise template fidelity from structured content.
Open source
Reference implementations of the patterns behind the writing — clean-room, tested, CI-green, and documented to run in a few minutes. Each one backs a specific piece on The Cloud Codex.
- agent-loop — a bounded agent loop with three hard exits: iteration cap, cost ceiling, and stall detection. TypeScript.
- inference-router — a provider-agnostic LLM client; swap backends with an env var, with a built-in smoke test. TypeScript.
- mcp-token-audit — measure the token cost of MCP server tool schemas and emit a defer-loading manifest. Python.
- token-baseline — measure a prompt/skill/command corpus, track reductions across snapshots, find duplicate files. Python.
Background
25 years in engineering leadership and platform work. Former CTO of a streaming platform (20+ person team, $4MM+ budget, 99.98% uptime, 100% YoY user growth, monolith-to-microservices migration). Six years as Senior Solutions Architect at AWS — migrating the workloads everyone said couldn't be moved, and a featured speaker on the AWS re:Think Podcast on cost optimization.
UC Berkeley Executive Education — Professional Certificate in Machine Learning and Artificial Intelligence (2025). Harvard Data Science Initiative — Agentic AI Intensive (December 2025).
Personal AI lab — NEXUS
Multi-machine personal R&D workspace prototyping patterns for client work. A NAS + Mac-mini + laptop topology with production services (TypeScript/Node), LaunchAgent automation, a custom Slack bot with auto-discovered slash commands, semantic search over conversation history (Ollama embeddings), self-hosted dashboards behind Cloudflare Tunnels, and integrations with Anthropic Claude, Plaid, Microsoft 365, Synology DSM, and GitHub. The test bed for prompt-contract patterns, agent orchestration, hot-loaded knowledge, and evaluation harnesses I later bring to enterprise engagements.
Writing
I write daily-ish about AI engineering, platform work, agent orchestration, and what the tokens actually cost at The Cloud Codex. Selected long-form, published at Presidio:
- What I Learned Building an Agentic AI Operating System
Practitioner voice on shipping internal AI tooling — enterprise AI is won on shared context across silos, not the newest model.
- The Six-Month AI Tool Half-Life: Designing for Replacement, Not Adoption
AI tools get swapped out within months — design for portability between platforms, not lock-in.
- Agentic Application Modernization Reality Check: You Can't AI Your Way Out of Technical Debt
Patterns applied to real modernization engagements — autonomous modernization is still years off.
Inquiries: mike@mpt.solutions · LinkedIn · résumé (PDF)