Work · A field report

Three private products. Native on iPhone, each also on the web.

These are real builds running on my own back-ends for family and private use — not public launches. The screenshots below come from the live apps. What they do, what they shipped with, and the engineering moves behind them.

Full-stack product · web + iOS

Household property tracker

items tracked: 248
categories: 22
rooms · web + iOS: 4
CPSC recall feed: 1

Private web + native iOS app for tracking household belongings — items, photos, serial numbers, receipts, warranties — with AI-assisted capture. Pulls CPSC recall data on a schedule and flags affected items.

Flask
React + esbuild
PostgreSQL
SwiftUI
Claude Vision

Home Inventory web app — dashboard view with 248 items across 22 categories, room filter chips, and a grid of item cards showing photos, brands, types, and model numbers.

Web

Home Inventory iOS app — dark-mode Outside room view with a search bar and a grid of item cards (Outdoor Furniture Set, Andersen Storm Door, Flame King Propane Tank, Smart Plug, Philips Hue bulb, Suncast deck box) and a floating tab bar for Items, Recalls, and Settings.

iOS · SwiftUI

01
Multi-item shelf scans: one photo, the model extracts every product visible — brand, model, and type — in a single Claude Vision call.
02
Receipt OCR with same-day-purchase suggestions — buy a new drill at Lowe's, the app proposes adding it within the hour.
03
CPSC recall watcher runs weekly, matches against owned items by brand + model, and flags affected entries on the dashboard.

Full-stack product · web + iOS + Android

Recipe library with retrieval-augmented chat

structured recipes: 190
passages searchable by chat: 1,000+
clients · web, iOS, Android: 3
Claude connector: MCP

A Next.js web app plus native iPhone (SwiftUI) and Android (Kotlin + Compose) apps over one FastAPI service. Turns recipe websites, cookbook PDFs and phone photos — handwritten family cards included — into structured recipes, and answers cooking questions with chat grounded in the recipes it retrieves.

FastAPI
Next.js
PostgreSQL
ChromaDB
SwiftUI
Kotlin + Compose
Claude Opus + Sonnet

Recipe iOS app — Chat tab in a dark theme with a copper flame mark, the headline 'What should we cook?' and three suggestion cards (Find recipes for what's in your pantry, Get cooking tips and substitutions, Plan your meals for the week) above an 'Ask about your recipes…' field; floating tab bar with Chat, Recipes, Pantry, To Make and Import.

iOS · Chat

Recipe iOS app — Recipes tab: a search field and meal filters (All, Breakfast, Lunch, Dinner, Dessert) above a two-column grid of photo cards, including Yellow Sheet Cake with Chocolate Frosting, Ultimate Flaky Buttermilk Biscuits (bookmarked), Ultimate Banana Bread, Twice-Baked Potatoes, Tortellini in Broth with Spinach and The Best Swedish Meatballs.

iOS · Library

Recipe iOS app — Cook Mode for Vanilla Wafer Cake: step 1 checked off and struck through with the ingredients it uses, step 2 highlighted as the current step, a 40:00 timer button on the 'Bake at 350° for 40 minutes' step, and a footer reading 'Step 2 of 8' with a progress bar, the next step, and Prev and Next buttons.

iOS · Cook mode

Recipe Android app — Chat: the question 'What can I make with chicken tonight' and an answer drawn from the household's own recipes, listing Fiesta Chicken Crockpot Meal and One-Pot Cajun Chicken Pasta with a short note on each; bottom navigation with Chat, Recipes, Pantry, To Make and Import.

Android · Chat

Recipe Android app — Recipes: a search field and meal filters (All, Breakfast, Lunch, Dinner, Dessert) above a two-column grid of photo cards, including Tortilla Casserole 1/2 Batch (bookmarked), Half-Loaf French Bread, Chicken Taco Salad, Hamburger Soup and Skillet Lasagna.

Android · Library

Recipe Android app — Cook Mode on step 1 of 4 of One-Pot Cajun Chicken Pasta: the step's instruction in large type, a 'For this step' list of just the three ingredients it uses (olive oil, chicken thighs, Cajun seasoning), a preview of the next step, and Mark done, Prev and Next buttons on a warm cream background.

Android · Cook mode

01
Grounded chat: vector + BM25 search fused and reranked, and every time, temperature or quantity in an answer has to come from a recipe it retrieved — no guessed cooking times.
02
A Claude connector (MCP): search and read recipes from Claude, and propose edits that change nothing until the person confirms them, with that person's own permissions.
03
Grocery receipts — photos or a PDF — are read, reviewed, and posted to the pantry in one step; the shopping list is what's planned minus what's already in stock.
04
Public potluck sign-up pages for work events: dishes by category, allergen notes, and a bot check on sign-up — the one part of the app that people without an account use.

Full-stack product · web + iOS

Theme-park dining concierge

parks & resorts indexed: 10
vector + BM25 fusion: RRF · k=60
rerank stage: cross-enc.
streaming chat: SSE

GPS-aware dining concierge across the major Orlando theme-park resorts. Hybrid retrieval over restaurant menus, reviews, and guest reports, with SSE streaming chat and blue-green deploys on a small ARM VM.

FastAPI
React + Vite
PostgreSQL
ChromaDB
SwiftUI
Claude Opus

Theme-park dining concierge web app — Chat tab welcome state. Left sidebar with park, area, and meal filters; main panel headlined 'Welcome' with suggested prompts like buffets at Animal Kingdom, mobile order tips, vegan options at EPCOT, and Deluxe Dining Plan value.

Web · Chat

Theme-park dining concierge web app — Browse view with a grid of restaurant cards spanning Magic Kingdom: Columbia Harbour House, Liberty Tree Tavern, Gaston's Tavern, Jungle Navigation Co. Skipper Canteen, Cinderella's Royal Table, Casey's Corner, Aloha Isle, Cheshire Cafe, Main Street Confectionery, Plaza Ice Cream Parlor, Sleepy Hollow, Sunshine Tree Terrace, Spring Roll Cart.

Web · Browse

Theme-park dining concierge web app — Map view of Magic Kingdom rendered with restaurant pins for table service, quick service, snacks, and lounges across the park's lands.

Web · Map

Theme-park dining iPhone app — Chat tab with park/area/meal filter chips at top, hero column with a Greek-temple mark, headline 'Where should we eat?', dining guide for Walt Disney World and Universal Orlando, suggestion cards for restaurant recommendations, DFB ratings, and menu browsing, and a Siri hint at the bottom.

iOS · Chat

Theme-park dining iPhone app — Browse view with Magic Kingdom park filter and a list of restaurant cards (Jungle Navigation Co. Skipper Canteen, Columbia Harbour House, Liberty Tree Tavern, Gaston's Tavern, Cinderella's Royal Table) each tagged with cuisine, service type, area, and a guest-rating badge.

iOS · Browse

Theme-park dining iPhone app — Map tab showing a Magic Kingdom map dotted with restaurant pins (Cinderella's Royal Table, Skipper Canteen, Sleepy Hollow, Liberty Tree Tavern, Cosmic Ray's, Plaza Restaurant, Sunshine Tree Terrace, Tony's Town Square) and filter chips for Table Service, Quick Service, Snack, and Lounge.

iOS · Map

01
GPS-first retrieval: when coordinates are available, location bypasses area-name matching entirely — closer beats louder.
02
Hybrid retrieval pipeline: vector + BM25 merged with reciprocal-rank fusion, then a cross-encoder reranker on the top-K.
03
SSE streaming responses with grounded citations: every restaurant fact links back to its source document.

On Apple platforms

iOS and macOS, end to end.

All three products above ship a SwiftUI iPhone app against a back-end I built myself. I work in Xcode across iOS and macOS — including signing, provisioning, TestFlight, and Apple Developer Program submission — so the same engineer who stands the server up is also the one shipping the native client. If your project needs a public iOS or macOS app, that pipeline already works on my machine.

SDKSwiftUI · Swift 5/6 · async/await · iOS 26+
AuthSign in with Apple, Google OAuth, bearer tokens
ToolingXcode, TestFlight, App Store Connect
NetSSE streaming, URLSession

Have something in mind?

If you're weighing a custom internal tool, a small public site, or a native iOS app to go with one — send a short description. You'll get an honest scoping reply, not a sales pitch.

Send an email