AI & automation

AI product artifacts for regulated, text-heavy workflows.

This page separates shipped product experience from AI prototype and product-thinking work. The goal is to show how I would define AI workflows, evaluation, trust controls, and measurable business outcomes without overstating what was productionized.

AI-enabled delivery

Portfolio launch automation system

Used AI as a product build partner to turn career knowledge, case-study material, resumes, and hiring goals into a deployed bilingual portfolio with an interactive fit agent.

Launched the first static Cloudflare version in less than a day, then iterated the UX over three days.
Structured the site as a living knowledge base for recruiters and hiring managers.

Webcoding prototype

Pipeline Recovery Agent

Built a public prototype that turns vague local-service requests into structured owner handoffs across plumbing, HVAC, roofing, lawn care, and cleaning workflows.

Used one reusable intake pattern across five route configurations.
Hosted the prototype on Cloudflare Pages so hiring teams can inspect it directly.

Open live demo

Side hustle / GTM

LeadFlow Studio website business

Built a live service-business MVP for small local businesses that need credible websites, lead capture, simple reporting, and a lightweight backend workflow.

Positioned the offer around preview-first risk reduction.
Uses Codex-assisted webcoding and Notion as a lightweight lead database and GTM learning loop.

Open live site

Prototype direction

AI underwriting assistant

Prototype workflow for turning broker chats, emails, and underwriting notes into summaries, extracted fields, recommended quote actions, and human review.

Classify intent from broker conversation signals.
Keep extracted fields editable, source-aware, and reviewable.

Product operations

Backlog and Jira workflow automation

Reduced refinement time by 25% through scoring improvements, clearer acceptance criteria, and a more structured prioritization workflow.

Turn ambiguous ideas into scoreable backlog items.
Use AI to draft acceptance criteria, edge cases, and visible tradeoffs before planning.

Artifact 00

AI-assisted portfolio build as a product workflow.

Live product artifact Open portfolio

Portfolio launch system

Built a bilingual portfolio from career notes and feedback, with case studies, resume routing, analytics, and a fit agent.

MVP< 1 day

Iteration3 days

AudienceHiring teams

01 Frame

Define audience, goal, proof points, and CTA paths.

02 Build

Generate structure, copy, UI variants, translation, and chatbot behavior.

03 Review

Correct facts, remove weak claims, and validate mobile and desktop UX.

04 Ship

Deploy to Cloudflare, add analytics, and iterate from live feedback.

What hiring teams can inspect

A live example of AI-enabled product execution.

AI accelerated the build, but PM accountability stayed human: goal framing, requirement clarity, content judgment, UX review, launch readiness, and iteration.

Product framing Converted an unclear career-brand problem into pages, user journeys, proof points, and CTAs.

AI orchestration Used AI to accelerate structure, UI iteration, bilingual content, chatbot prompts, and deployment checks.

Inspectable output Shipped a static site with case studies, resume variants, role-fit scoring, analytics, and live links.

Read case studies Ask Fit Agent

Artifact 01

Pipeline Recovery Agent live prototype.

Pipeline Recovery Agent Live demo

Routes5

Flow4 steps

BuildAI-coded

HostCloudflare

Customer signal

Emergency leak, no heat, storm damage, cleanup request, or recurring lawn-care quote.

Agent handoff

Service category, urgency, location, timing, recommended reply, and next action for the owner.

What it demonstrates

This prototype shows how I use AI and webcoding to move from product idea to inspectable workflow quickly. The point is not to claim a production business result; it is to show requirement framing, reusable workflow design, live demo quality, and AI-assisted execution.

Problem framingRecover missed leads when local-service owners cannot answer every call or incomplete request.

Reusable designKeep one renderer and use route configuration for different service categories and urgency logic.

Inspectable outputGive hiring teams a live URL they can click through instead of only reading a slide.

Open Pipeline Recovery Agent

Artifact 02

LeadFlow Studio as an AI-assisted webcoding business.

LeadFlow Studio Live MVP

OfferWebsite + leads

ResearchInterviews

BackendNotion

BuildCodex

Small-business pain

Website cost feels risky, DIY takes time, and existing pages often do not capture requests or prove what is working.

Productized solution

Preview-first website, quote form, lead table, owner alerts, and simple reporting that can repeat across local-service niches.

Why it belongs here

LeadFlow Studio is a side-business MVP that shows practical AI-enabled product execution. It combines customer discovery, offer design, Codex-assisted webcoding, Cloudflare deployment, and a Notion-backed backend concept for leads, research notes, surveys, and GTM learning.

0 to 1Validate the problem through owner interviews, website audits, surveys, and preview-first pilots.

1 to scaleTurn repeated pain points into templates, package tiers, Notion schemas, monthly reports, and vertical-specific GTM.

AdvantageUse AI webcoding to reduce build time and Notion to keep the backend simple before investing in custom infrastructure.

Open LeadFlow Studio

Artifact 03

Human-in-the-loop underwriting assistant.

Broker thread summary Review required

Broker

Client changed travel dates and needs updated coverage options before payment.

Assistant

Intent: quote update. Missing: departure date confirmation. Recommended action: request date confirmation, then refresh quote.

IntentQuote update

RiskMissing date

ActionRequest info

OwnerUnderwriter

Product decisions

The assistant should not silently update a policy or quote. It should reduce reading and extraction work, then ask the underwriter to approve, edit, or reject the recommendation.

Source contextShow which message or note supports each extracted field.

Editable outputLet users correct the summary, field values, and recommended action.

EscalationRoute low-confidence or compliance-sensitive cases to manual review.

Artifact 04

Evaluation rubric for an AI workflow assistant.

Accuracy

Key-field extraction

Are dates, plan changes, traveler details, and quote intent extracted correctly from source text?

Completeness

Missing information

Does the assistant identify required fields that are absent before recommending a quote action?

Trust

Source traceability

Can the underwriter see why the assistant made a recommendation and where the evidence came from?

Risk

Hallucination control

Does the workflow prevent unsupported facts from moving into quote or policy updates?

Adoption

Human acceptance

How often do underwriters accept, edit, reject, or escalate the recommendation?

Business impact

Cycle-time reduction

Does the assistant reduce review time without increasing quality or compliance risk?

Framework

The AI product pattern I want hiring teams to see.

1Capture signals

Tickets, chats, product events, VoC, QA defects, and operational notes.

2Extract meaning

Intent, risk, priority, missing data, product rule conflicts, and customer friction.

3Recommend action

Next step, owner, validation path, acceptance criteria, or escalation path.

4Measure impact

Cycle time, conversion, cost, quality, satisfaction, and operational load.

Artifact 05

Backlog automation pattern for product teams.

Raw input

Customer feedback, stakeholder request, Jira note, QA defect, analytics signal.

AI draft

User story, acceptance criteria, edge cases, dependencies, and open questions.

PM review

Clarify tradeoffs, remove unsupported assumptions, score priority, and align scope.

Sprint-ready

Refined ticket with owner, success metric, validation path, and release risk.