AI innovations,
across industry.
A portfolio of AI applications and agent architectures I've designed across industries, some shipped to production, others mapped out as blueprints. Organized by the business problem they solve, not by the model that happened to be in fashion.
The interesting question was never "where can we add AI?" It's which problem in this industry is really a reading, reasoning and form-filling problem in disguise, and then designing the system that solves it.
What I've built,
and what I'd build.
A mix of systems shipped to production and agent architectures designed for the problem, each tied to the industry it serves and the business outcome it moves.
An agent team for tax & filings
A graph of narrow AI specialists, tax-law, accounting-data, deductions, compilation, a critical thinker, an auditor and an orchestrator, that reads the data, computes against the rules, and drafts the filings, with the accountant accountable at the end.
- →One tax-law agent per section: personal, corporate, trusts
- →A deductions agent that hunts for legitimate savings
- →A graph, not a swarm, so every conclusion is traceable
- →Human signs off on a summary of the reasoning, not raw data
Margin-finding agents for fulfillment
In a commodity market the price is fixed, so profit lives in the cost of fulfilling each order. A graph of agents sources every line, across owned inventory and drop-ship suppliers, picks the cheapest carrier that still meets the promise, and holds every order above a margin floor.
- →Sourcing agents across owned stock and drop-ship suppliers
- →Carrier-rate agents pick the cheapest path that meets the promise
- →A landed-cost agent computes true margin per fulfillment plan
- →A margin guardrail blocks any order that would ship at a loss
Turning public healthcare data into answers
The government publishes an enormous amount of healthcare data that almost nobody can actually use. I built the thing that makes it usable: ask a question the way you'd say it out loud, and get a clear answer in seconds, the kind that used to mean hiring analysts and waiting weeks. It's the same instinct as the agent work, do the hard part once, up front, so every answer after that is quick and cheap.
- →All of it gathered into one place, finally lined up so it can be compared
- →Stays current on its own as new data is published
- →Ask in plain English, no specialists or special skills
- →Powers clients' own dashboards and reports
A marketing department that runs on one person
Search Console and Analytics already know which queries you nearly rank for, which pages lost position last month, and which intent nothing on the site answers. Almost nobody mines them that way, because doing it properly is a full-time job nobody has. This system does it every day, watches what competitors publish alongside it, then drafts the content, generates the social posts and the imagery, and schedules the calendar. The marketer stops producing and starts directing.
- →Search Console and Analytics read continuously: queries sitting on the edge of page one, pages losing position, intent with no answer on the site
- →A competitor watch that reports what they published, what they quietly changed, and what it earned them
- →A content agent that drafts against the specific gap it found, in your voice, with the brief attached so an editor can argue with it
- →Social copy and channel-native imagery generated per platform, then scheduled into a calendar a human approves before anything publishes
- →One marketing expert directs the system and owns what ships. The throughput belongs to a team.
Twenty-one agents reading one video
Watchly, the kids' YouTube player I own and run, works on one rule: a parent approves every video before a child can watch it. That rule only holds if approving is quick, and a title and a thumbnail tell a parent almost nothing. So the system watches first. It reads the transcript, samples frames across the runtime, and hands that evidence to twenty-one agents, one for each theme a child shouldn't run into. Each returns a severity and a written opinion citing the moment it reacted to. The parent still decides. They decide with the video already read.
- →Evidence before analysis: the transcript segmented with timestamps, plus frames sampled across the runtime, because a video can be visually wrong with perfectly clean audio and the reverse is just as common
- →One agent per theme, twenty-one of them, each carrying its own rubric and its own age thresholds. Profanity and self-harm have no business on the same scale
- →Narrow agents instead of one prompt asking twenty-one questions. A single prompt spreads its attention across all of them and fails quietly, which is the worst failure mode a safety system can have
- →Every agent returns a ranked severity and a written opinion that cites the timestamp behind it, so the reasoning is auditable rather than a number to trust
- →An adjudicator reconciles the overlaps, weighing cartoon violence differently from documentary violence, and normalizes severities so ranks from different themes are comparable
- →A cheap triage pass runs first, so the full twenty-one-agent fan-out is only paid for on videos that warrant it
- →One row per theme per video in the database, opinion and cited evidence included, so a parent reads an argument instead of a score
- →Evaluation per agent rather than on the blended result. A single theme agent drifting is exactly the failure an overall safety score would hide
From problem to system.
Start from the problem
I start from the business outcome, not the model. The first question is which work in an industry is really reading, reasoning and computation in disguise, and where AI is just a distraction.
Design the agent system
Most of these are agent architectures: narrow specialists organized under a hierarchy you can trace, designed so the output is something a professional can defend, not a confident guess.
Ship where it counts
Where a design goes to production, it ships with evaluation, observability and cost control, and a human kept at the point where judgment and accountability belong.
What teams ask about this work.
Have you shipped all of these, or are some designs?
Both, and mostly shipped. The marketing system, the order-fulfillment agents and Watchly's video review run in production today, and the healthcare data tool is live in client use. The tax-and-filings architecture is documented in its write-up and gets adapted per engagement.
Can you adapt one of these to my industry?
Usually, yes. The pattern, narrow specialist agents under a traceable hierarchy with a human accountable at the end, transfers across domains. The work is mapping it to your data, your rules, and your regulatory reality.
Do you build, or only design?
Both. I work hands-on, architecting and, where it makes sense, building and shipping the system with evaluation, observability and cost control. Some engagements stop at a design and a roadmap your team executes.
How do you keep these reliable in a regulated industry?
By organizing agents as a graph rather than a swarm, so every conclusion is traceable, and by keeping a human at the end of the process to review the reasoning, not just the result.
Have a problem
that AI could solve?
Tell me the industry and the outcome you're after. I'll tell you straight whether AI is the right tool, and what it would take to design and ship it.