Builder

An IWF-compliant Olympic weightlifting competition simulator, built end-to-end as a data platform on ~7,000 real USAW competition records.

A full gym-management platform for a real weightlifting gym — public marketing site, athlete portal, and an admin back office with a block-periodization program builder — built on Next.js, Prisma, and NeonDB.

A camera-first velocity-based training PWA that turns any phone into a bar-speed sensor — with AI autoregulation, BLE hardware support, and a live coaching dashboard.

A professional weightlifting team-management platform — coaches build and assign periodized programs, athletes log workouts mobile-first, and progression is tracked end to end.

A research-grade dataset and interactive dashboard covering 770 USAW-affiliated weightlifting clubs across all 50 states — with public pricing for 331 (43%) — built entirely from free public web research.

The story & how it’s built

Because I compete in USAW weightlifting and wanted to know what coaching actually costs across the country, I kept hitting the fact that most clubs deliberately hide pricing behind "fill out our form" lead-capture gates, and nobody has ever published the actual distribution, so I built a dataset + dashboard that documents which clubs publish pricing, what they charge, and — just as honestly — which don't, that turns a market's best-kept number into an open dataset. It taught me that "we couldn't find it" is data too: the 57% without pricing is the finding.

View on GitHub ↗

Collects pricing data from 20+ weightlifting equipment retailers — barbells, plates, racks, belts, apparel, and shoes — and makes it publicly browsable.

The story & how it’s built

Because I train and compete in Olympic weightlifting and buy gear regularly, I kept hitting the fact that equipment prices vary wildly by retailer and there's no single place to compare, so I built a scraper that polls 20+ retailers and collects of barbell, plate, rack, belt, apparel, and shoe prices into one database, that lets me see who has the best price on a Rogue Ohio bar this week without checking six sites. It taught me that the most valuable public dataset is the one nobody else bothered to build because it's "just scraping."

View on GitHub ↗

An all-in-one wedding platform that replaces HoneyBook + Pixieset + Calendly + DocuSign in one tenant — CRM, galleries, invoicing, timeline, contracts, print store, and partial payments.

A mobile-first PWA that photographs a physical event contract or run-of-show sheet and extracts the details into a searchable database with Claude Vision — so "where's that paperwork?" stops being a question I ask mid-event.

A sales negotiation co-pilot for wedding vendors — import a call transcript and get a report card on what you said, what the client signaled, and what to try next.

The backend that collapses a wedding vendor's 4–5 tool stack — CRM, galleries/delivery, file storage, email — into one multi-org service, with vendors bringing their own Dropbox, their own branded email templates, and their own gear inventory.

A planner that generates and formats the day-of run-of-show timeline for a wedding — the second, full rebuild of a tool I first shipped as a simple public version.

A DJ suggestion engine that reads the room and suggests next tracks, transitions, and crowd-management moves — built as a FastAPI + mobile PWA.

The story & how it’s built

Because I DJ weddings and events every weekend, I kept hitting the fact that the best next-track decision is obvious in hindsight but hard in the moment — especially when you're managing a mic and a playlist, so I built a co-pilot that knows the current track, the room energy, and the genre map, and suggests what comes next, that gives me one less thing to think about when I'm running sound. It taught me that the most useful tool is the one that takes a 1-second decision down to zero seconds — not the one that generates perfect set lists.

View on GitHub ↗

A gated, read-only web UI over the nationwide wedding vendor dataset — browse 1,734 vendors, explore 32k+ reviews, and ask plain-English questions with text-to-SQL.

The story & how it’s built

Because I built the scraper pipeline that collected TheKnot + Zola data, I kept hitting the fact that a Postgres database is useless to non-technical people who want to browse it — "just write SQL" isn't an answer for an office, so I built a gated snapshot app: Postgres → SQLite export → read-only FastAPI UI with filters and an ask-the-data box, that lets anyone on the team explore vendors, reviews, and pricing without touching a query. It taught me that the last mile of a data project is an interface, not a pipeline.

The operating system for a two-person photo/video business — projects, invoicing, contracts, proposals, and CRM — built to replace the pile of spreadsheets and HoneyBook that were running the studio.

The commercial front door for LA Media — a portraits/sports/weddings photography site with a cart and a client portal — built to turn "shooting favors for friends" into a packaged offering people can actually book and pay for.

A photography lighting tracker that tells you exactly when and how to shoot — astronomical sun/moon math plus live weather feeding AI camera-setting recommendations.

A lightweight macOS desktop app that copies photo selects out of a shoot folder — by pasted shot numbers or in bulk — so culling a gallery stops meaning dragging files by hand.

A B2B lead discovery and enrichment pipeline — SMTP-verify emails, deep-crawl company websites, and render the results as a CRM view.

A Playwright scraper that watches every lot at Earl's live auctions, tracks each bid over time, looks up the retail price, and tells me the actual dollars-and-percent savings — so I know which lots are a real deal before I bid.

Automated price collection across fourteen Indianapolis-area grocery retailers — each one reverse-engineered with the specific technique that store's site actually requires, including a Trader Joe's scraper that beats an Akamai WAF — into a PostgreSQL database for trend tracking and cross-retailer comparison.

An AI-forecaster trading bot for Polymarket that finds mispriced prediction markets and trades the edge — with a real risk manager and the discipline to stay in paper-trade mode until the data earns real money.

A weekend flight-deal platform that beats Skyscanner's PerimeterX anti-bot — mobile-API PX cookie solving + residential proxies feeding a self-hosted deal tracker with price history and 200+ destination coverage.

A production-grade, Dockerized scraper that started as a single-site vendor collector and grew into a nationwide, 3-marketplace vendor-and-review intelligence pipeline with its own sentiment-analysis dashboard — a written engineering analysis, a test suite, and pre-commit gates behind it the whole way.

Snap a photo of a thrift find and get a HOT/GOOD/PASS deal score in under 20 seconds — Claude Vision identifies the item, live eBay comps price it, and the app tells me whether the margin is real before I ever put it in the cart.

A multi-source wedding vendor pricing research tool that scrapes DJ, coordinator, and photobooth pricing from TheKnot, Zola, and WeddingWire — surfacing market rates for a data-advantaged sales approach.

A standalone eBay Marketplace webhook service — challenge-response verification, ECDSA-signed deletion notifications, and a status dashboard — the compliance plumbing every eBay marketplace app must have.

A tool that pulls live Public Storage unit pricing across locations and a Telegram bot that uses Claude to tell me whether the current rate is actually a good deal.

The story & how it’s built

Because I was pricing self-storage units, I kept hitting the fact that Public Storage rates fluctuate constantly and there's no easy way to know if right now is a good time to book, so I built an aggregator that scrapes rates across locations plus a Claude-powered Telegram bot that judges the deal, that turns "is this a good rate?" into an on-demand answer in my chat app. It taught me the reusable pattern: scrape a market, then let an AI make the buy/skip call — the same shape I later scaled up in [POLYBOT](polybot.md).

A price tracker and listing finder for Nike Romaleos 2 weightlifting shoes — live eBay Browse API search, US-size extraction, smart watches, and Discord alerts.

The story & how it’s built

Because I compete in weightlifting and Romaleos 2s are the grail shoe — discontinued, so the only market is eBay resellers, I kept hitting the fact that good pairs appear randomly, sell fast, and every listing hides its size inside a messy title, so I built a tracker that polls the eBay Browse API, extracts the US size from each title, and pings Discord when a watch matches, that turns "check eBay obsessively" into a filter and a push notification. It taught me that the hard part of deal-hunting isn't finding listings — it's structuring them (size, condition, freshness) so an alert can be trusted.

View on GitHub ↗

A GitHub-style contribution grid for personal productivity — track daily tasks, build streaks, and see the whole year at a glance.

A local-first voice journal — talk, get it transcribed and optionally summarized, track habits and mood — built so my thoughts live on my device, not on someone's server.

A self-hosted dashboard that pulls my real money — Fidelity, Chase, Citi, PayPal, and Venmo — into one net-worth view, so I stop logging into five sites to answer one question.

A pet-care app built for non-technical pet owners — medical records, breed reference, weight tracking, and a printable vet report, all in one clean PWA.

A cannabis strain catalog with genetic lineage trees, terpene profiles, effect browsing, and a dispensary map — 4,800 strains enriched from a four-source data pipeline with 59K genetic relationships.

A social wine tracking & discovery app — snap a bottle *or a glass*, tag where you're drinking on an interactive map, rate it, and see what your community is uncorking nearby, backed by a ~50,000-winery location database.

An interactive 3D maintenance manual for my own 2011 Chevrolet Traverse — click any part of the model to get service procedures, troubleshooting, and AI-assisted diagnosis.

A Logos-style Bible study app — commentary, word study, and an AI study assistant — built because the established software costs a fortune.

The story & how it’s built

Because I wanted to study Scripture deeply, I kept hitting the fact that the serious tools (Logos and the like) are expensive and heavy, so I built my own study app with commentary, word study, and an AI assistant, that gives me the depth I wanted without the price tag, It taught me a full-stack build with real theological data ingestion — FastAPI backend, structured study data, and an AI layer over it.

View on GitHub ↗

A civic-data aggregator that grew from a single city into a 3-region platform — Fishers, Indianapolis public safety, and Hamilton County GIS — unifying council votes, procurement bids, campaign finance, zoning, court records, crime/crash/use-of-force data, 311 requests, and county parcel/school/park layers into one searchable, alerting dashboard.

A lost-and-found pet platform for the Indianapolis metro that scrapes five separate pet-listing sources — each one reverse-engineered differently — geocodes them onto one map, and matches lost dogs against found reports.

A self-hosted market-intelligence agent and automation layer running on a personal VPS — proof that Alex builds infrastructure, not just apps.

A zero-budget commodities and macro-cycle dashboard tracking 93 data series across 8 metals, 15 stocks, 28 macro indicators, COT, LME, and ERCOT — all from free sources.

A live command center for every project in this portfolio — real-time GitHub stats over Server-Sent Events, an agent-facing JSON API multiple Claude Code sessions can safely read and write concurrently, and human login + email 2FA gating the view — the tool built to manage the tools.

A personal penetration-testing framework, custom MCP servers for Kali Linux and Metasploit, and ~18 months of hands-on bug-bounty work — 78 verified findings across 21 targets on HackerOne and public VDPs — applied security R&D, not CTF hobby work.

A structured knowledge base of LLM API pricing — 216 price points across 11 providers with historical trends (2020–2026), free-model availability, and market context — plus a generated dashboard and report.

The story & how it’s built

Because I run a fleet of AI-powered projects on OpenRouter and needed to know what inference actually costs over time, I kept hitting the fact that pricing pages are marketing surfaces — per-token rates change silently, free tiers come and go, and nobody keeps the history, so I built a canonical JSONL dataset (one file per provider) with every price change timestamped, plus a generated dashboard and report, that gives the portfolio's model-selection decisions a data foundation. It taught me that "what does GPT-4 cost" is a time-series question, and everyone else is only looking at today's number.

View on GitHub ↗

A self-run home stack — NAS, mesh VPN, backups, and a self-hosted multi-account AI proxy — debugged at the OS and network level.

The story & how it’s built

Because I run my own tools and wanted them available and backed up everywhere without renting someone's cloud, I kept hitting the low-level failures nobody hands you a GUI for — stale proxies, symlink path mismatches, services racing to start before the network was up, so I built and hardened a home infrastructure stack I own end to end, that keeps my data, dev environment, and AI tooling running reliably. It taught me to diagnose problems at the OS and network layer, not just the app layer.

A research companion to the Infra Metals Dashboard — silver supply-side modeling, decades-long macro series, and the analytical write-ups that turn raw data into investment decisions.