Projected reduction in analyst time-to-insight
Led a governed enterprise GenAI product from use-case definition through architecture review, delivery planning, and an ROI model tied to $60M in annual revenue opportunities.
AI Strategy · Governance · Hands-on Delivery
I'm an AI strategy and governance leader with 10+ years bridging executive priorities and technical delivery. I design the operating model, align the people who own the risk, and stay hands-on enough to prove what works.
The public writing shows how I think. These anonymized outcomes show the scale of work behind it: AI strategy, governance, product leadership, and operating-model change inside large organizations.
Led a governed enterprise GenAI product from use-case definition through architecture review, delivery planning, and an ROI model tied to $60M in annual revenue opportunities.
Led data-catalog integration, lineage, stewardship, training, and change management so the governance platform became part of how teams worked.
Mapped fragmented enterprise processes into a reusable operating model that made delivery knowledge easier to find, teach, and apply.
Client work is summarized and anonymized. See how I approach AI governance and data strategy →
I stay hands-on outside enterprise work by shipping small products, running AI infrastructure, and publishing the checks and tradeoffs behind the build.
Guides, templates, and tools for the person who ends up coordinating the group trip. Built around itineraries, decisions, and reducing planning chaos.
A free, no-signup party game site — multiplayer Wordle, Connections, Ito and more, played from your phones. Built solo and audited with a fleet of AI agents before launch.
A newborn sleep-tracking app inspired by the gap between paid baby apps and the simple logging workflow I wanted at 3 a.m.
An ongoing community for people learning AI by building real workflows and small apps, not just collecting prompts or watching demos.
Three threads, each written as a dependency chain rather than a feed. The numbering is the argument: later pieces assume the earlier ones. If you only read one from a thread, read the one marked start here.
Access control is the visible layer. These are the layers underneath and above it that decide whether it actually holds.
A real multi-machine fleet running scheduled agent work. Built to test enterprise AI claims against hardware I actually control.
Products and pipelines built solo, written up with the parts that did not work left in.
Browse by topic: AI & Building · Data & AI Governance · Tech & Tools · Life & Travel · Book Notes
Two dashboards, two revenue numbers, both defensible. The metric definition is a governed asset — and in most estates it lives in the one place governance never reaches: inside the BI tool.
A contract says what a column is; an SLA says whether it showed up. The four dimensions worth promising, why detection is cheap and paging is expensive, and how to measure what the consumer actually feels.
A real bake-off on this site's own posts: one written spec, two agent stacks, and a blind third-party judge that scored them 76-75 — a tie the process differences explain better than the number does.
A constraints-first system for AI trip planning: route skeletons before day-by-days, a verify step for every bookable claim, and the places AI still fails — from 34 days in Italy with a toddler.
Contracts, classification, and access control all assume you can trace a column backward to its source and forward to what depends on it. How lineage actually gets captured, why column-level is the threshold that makes it useful, and why the graph is a lower bound.
Access control and AI-agent governance both assume a classification layer that's accurate and current. Here's how you actually discover, tag, and maintain it — the taxonomy, the three discovery signals, and why it goes stale.
Downstream data-quality tests catch damage that already happened. A data contract moves the check to the boundary the producer controls — schema, semantics, quality thresholds, and change policy, enforced and blocking.
Score customers on Recency, Frequency, and Monetary value in SQL, then let a language model name the segments and draft the playbooks — with a hard line about which side owns the numbers.
I gave an autonomous coding agent one goal: make my shelved funnel-mapping app launch-ready. What it shipped, how I verified it, and where the product goes next.
How Search Console data, 301 redirects, refreshed titles, and AI-assisted review turned an old WordPress footprint into a cleaner static-site loop.
How I built a free, no-signup multiplayer party game site — and used a fleet of AI agents to audit it for bugs and security holes before launch.
Why paying for Huckleberry made me want to build a simpler tracker, and what that says about learning AI by making useful apps for your own life.
Why a single AI-drafted report can't be trusted on its own, and the adversarial, multi-model verification pattern that catches fabricated numbers before they ship.
Enterprises are deploying AI agents faster than their access-control models can absorb. A least-privilege framework for non-human identities: service identities, scoped tokens, and autonomy tiers mapped to data sensitivity.
Grant sprawl is a math problem — teams × roles × contractors. ABAC only fixes it if classification hygiene comes first. Practitioner notes on what changes, what doesn't, and where contextual policy is headed.
Turning the GA4 BigQuery export into customer intelligence — schema essentials, sessionization, identity stitching, and RFM, LTV, and cohort marts. The practitioner patterns.
llama.cpp vs Ollama vs MLX, quantization tiers, hardware sizing, model roles, and how to scale from one machine to a failover fleet. The practical guide.
How I moved from a model-centric approach to a system architecture — routing, memory, privacy boundaries, machine roles, and the governance layer that makes it durable.
How real capability scorecards get weighted before vendors ever demo, what belongs in a governance-tool TCO model beyond the license line, and how to cut a 20-vendor field to two without running 20 demos.
Org-chart steward assignments, unfunded curation work, definition debates re-fought as catalog edit wars, and the vanity metric that was lying to us — anonymized from an enterprise Alation rollout.
RAG on an ungoverned data estate rediscovers stale definitions, homonym collisions, and answers with no lineage. The boring governance artifacts were retrieval infrastructure all along.
Everyone blames the MDM platform for dirty customer records — the real damage happens upstream at intake: free-text fields, invisible validation logic, and the portal redesign that actually fixed it.
How I use agents, scheduled checks, documentation, model testing, and DataOps opportunity discovery to turn messy inputs into structured next actions.
Google AI Studio for scripts, Vids for presentations, Veo 3 for B-roll, Whisk for thumbnails. A practical walkthrough of the full AI-powered YouTube production workflow.
NotebookLM now generates full AI explainer videos from your sources. What the output actually looks like, when it's useful, and what it means for content creators.
Tokens, RAG, hallucinations, and why private data is where enterprise AI actually gets interesting. A plain-language breakdown of how large language models work.
Hands-on comparison of leading AI app builders using the same prompt. Which one actually ships a usable product in a single pass?
Three machines, a custom routing layer, and 24/7 local inference. The architecture behind running your own LLM fleet on consumer hardware — no cloud APIs required.
The gap between AI hype and AI that delivers value in enterprise environments. Patterns that work, patterns that stall, and the GTM playbook for data teams.
How I use Google AI Studio, automated outlines, and AI-assisted editing to keep shipping content with almost zero free time. The workflow that turns 20 stolen minutes into a finished draft.
Comfort graveyards, the 5 Whys technique, and skill stacking. A framework for confronting the internal barriers that actually hold you back.
Jim Collins' framework mapped to real career decisions. Confronting brutal facts, disciplined execution, and finding the intersection of strengths, passion, and economic value.
The tools and pipeline I use to turn long-form recordings into clips. AI-powered editing that handles the tedious parts so you can focus on the message.
What happens when you have the stable career, the remote job, the flexibility — and realize something's still missing. Building feedback loops, creative output, and purpose outside the day job.
Window Trail vistas, Rio Grande hot springs, and dark-sky stargazing across nearly a million acres. Notes from the drive and the trails.
The real logistics of a month in Italy with a toddler. Route, gear, flights, trains, ferries, and everything I'd change.
The first trip was the golden route. The second went deeper. What I'd tell first-timers vs repeat visitors.
The baby gear that actually earned its place on a 34-day trip through Italy — what to bring, what to leave, and what to buy when you land.
The exact setup I used to work 34 days across Italy with a baby in tow — laptop, connectivity, power, capture, and backup.
Install, audio setup with BlackHole, scene config, 1080p/4K recording settings, and how M-series MacBooks perform.
Separating mic, desktop, and game audio in OBS with BlackHole — and why separate tracks make editing so much easier.
The Fusion node blur technique, step by step — works in the free version, covers tracking moving faces for privacy and compliance.
Beyond the meme: obsessive preparation, fearlessness as a practice, and the 1% compounding thesis applied to career and growth.
What I’m building and learning — AI experiments, data work, apps, and the occasional trip. No fixed schedule, no fluff.
No spam, unsubscribe anytime.
You're in — check your inbox for a welcome note.
I'm Alan Nafarrete — an AI strategy and governance leader. I help large organizations move from scattered experiments to working operating models: clear ownership, safe AI usage, trusted data, measurable value, and delivery teams that can execute.
I work as a player-coach. I can facilitate the executive and risk decisions, then get close enough to the data, architecture, and workflow to keep the strategy grounded. Outside client work, I run a local AI inference fleet, ship small products, and publish what the verification process teaches me. I also run Blendlogic Tech for real-world MacBook performance testing.
The AI Builders Community is the free, ongoing place to compare notes on local infrastructure, agent workflows, and useful software. No course funnel — just build notes and people shipping.