I scope fuzzy problems
None of these products started as a ticket. Each one started as a vague need I turned into a concrete build plan: features cut, constraints named, first version defined.
P3
Palumbo's
Project
Portfolio
Forward Deployed Engineer
I do the job forward deployed engineers are hired to do, and I have the running software to prove it. I take a fuzzy problem, scope it into something buildable, build it inside the stack end to end, ship it to production, and then keep operating it. Below is that arc, repeated across a fleet of AI products I run today.
What the fleet proves
I build inside a stack end to end
Frontend, typed Worker API, D1 schema and migrations, auth, payments, and the deploy pipeline. I work every layer, not a slice of one.
I own deploy and operations
Every product deploys on git push and stays up because I run it: migrations, monitoring, incident fixes, and cost control are mine.
Case studies
Four of the products, told the way the work actually went.
Lost Pet Radar
lostpetradar.com- Problem
- When a pet goes missing, reports scatter across posters and feeds, and the first hours matter most.
- Embedded
- I scoped it around one constraint: a panicked person on a phone. Location-first reporting, no account friction, alerts that reach neighbors.
- Built
- Interactive Leaflet map with search-radius rings, photo uploads on R2, area alert subscriptions, passwordless OTP auth, and Workers AI for pet identification and content moderation.
- Shipped
- Live on web and iOS, deployed from the same git-push pipeline as the rest of the fleet.
- Running
- Alerts fire on schedule and moderation runs on every report. I operate it daily.
WriteMyCard AI
writemycardai.com- Problem
- People stall on the blank inside of a greeting card.
- Embedded
- I scoped the writing flow first: occasion in, polished draft out, with edits that never lose the user's own words.
- Built
- LLM drafting with alternates, inline editing with a full change trace, version history, and one-time token billing through Stripe with idempotent crediting.
- Shipped
- Launched as a paid product with Turnstile bot protection in front of every expensive call.
- Running
- Billing, drafting, and abuse controls all run in production today.
H2T2H
h2t2h.com- Problem
- Tense conversations escalate fastest in text, exactly where people have the least coaching.
- Embedded
- I grounded the product in a ten-principle communication rubric before writing code, so the AI had a standard to coach against.
- Built
- An AI coach that rewrites tense messages with per-change annotations tied to the rubric, tone controls, and soft, neutral, or firm reply options. Subscriptions sync through Stripe webhooks.
- Shipped
- Launched as a subscription product on the fleet's shared auth and deploy pipeline.
- Running
- Webhook-synced billing and coaching run in production today.
Vibe Helping
vibehelping.latentedge.io- Problem
- Feedback on live websites travels as screenshots, and screenshots lose the element they were about.
- Embedded
- I scoped it to where reviewers already are: the page itself, with comments anchored to the exact element.
- Built
- A Manifest V3 browser extension and web app sharing one Workers API: threaded element-level comments, team invites, completion tracking, JWT auth, and exports formatted as AI-ready prompts for agent handoff.
- Shipped
- Extension and web app shipped together against the same API.
- Running
- Teams use it on live sites today, and its exports feed my own agent workflows.
Also built and running: Shout Out Classy, a creative brand platform with a full Stripe cart-to-checkout pipeline, and The Grateful List, a gratitude journal with a public community feed. The full map of projects, platform, and skills is on the home page.
The operating layer
The fleet shares a platform I also built and run: a common scaffold, in-house auth and email packages, and an agent-driven roadmap system that opens pull requests across every project. That is the part of the job that never shows up in a demo: keeping many production systems consistent, cheap, and shippable at once.