Builder · Philadelphia, PA
I build work zone safety software, an AI grading tool for teachers, a competitive intelligence product, and a hundred browser utilities. I also build two iPhone games. I started in 2010 on a machine shop floor, measuring parts that had to be right. I still build the thing, then find out whether it actually works.
Machine shop, then battery research, then pharma analytics, then enterprise people data. A product was shipping next to the day job for most of those sixteen years.
I designed and shipped these myself. Each card says what you can open and what you cannot.
An open source intelligence (OSINT) watch floor for more than one theater. Every event is sourced and confidence tagged, with a live claim layer from the Global Database of Events, Language, and Tone (GDELT).
The license is MIT. It is not a targeting product. The map and the source are both public.
An Orion window replay of the Artemis II lunar flyby. It uses NASA's real lunar flyby photos and the crew's own voice recordings, timed to the second, plus a GoPro view out Window 3 as the Moon passes. No AI imagery and no fake sound. The repo includes an open source Python command line tool that searches and downloads the Artemis II Planetary Data System archive.
An AI competitive intelligence platform for early stage biotech. It tracked competitor movement inside a company's therapeutic area and modality, along with investor and funding shifts, partnership announcements, regulatory milestones, clinical data, mergers, pipeline changes, and signals in papers and patents. Retrieval runs over a vector database. Large language models (LLMs) curate and summarize.
I built it inside a pharmaceutical marketing agency so the team could stop tracking competitors by hand. It worked well enough to get a name, a logo, and a price of $9,000 a year, then launched to the biotech clients of an affiliated life sciences advisory firm. Sixteen pharma brands were modeled in it. Each one was mapped to its disease state, therapeutic landscape, and named competitive set.
Construction work zone safety software. Safety representatives author Internal Traffic Control Plans (ITCP) on a canvas. They place equipment and worker zones, and the app flags struck by conflicts before anyone is on the road. A plan publishes to a QR code so the crew can acknowledge it on a phone at the job site.
Struck by incidents are a leading cause of construction fatalities. The plans meant to prevent them are still drawn in marker on a tailgate. I want a safety rep to author and publish a real plan in under thirty minutes.
Two products for opposite ends of the pest control industry. Bugsy is a free consumer app. A photo of droppings, tracks, or damage goes through image recognition and comes back as a pest identification with a confidence score, then either a do it yourself treatment path or a route to a qualified local pro. The PestTech Management Portal is cloud software for pest control companies. It covers scheduling, customer chat, marketing automation, analytics, and payments.
I am co-founder and technical lead on a four person founding team, with a pest control operator, a marketing director, and a business development lead. The first markets are Oklahoma, Colorado, and Pennsylvania. Consumers often guess wrong before they call anyone, and the companies they call are running on fragmented tools.
An AI paper review tool for teachers. A teacher uploads student writing and gets structured feedback that follows the rubric. The grading pass drops from hours to minutes, and the teacher still makes every judgment.
I built and debugged the production stack across three services on my own. It has since been assessed as a sellable asset on acquisition marketplaces.
112 browser calculators and utilities across 8 categories, roughly 170 URLs. Every tool runs in the browser. There are no uploads, no accounts, and no display ads.
This is an experiment I am running in public. Impressions jumped roughly a hundredfold after a crawlability fix, and clicks stayed flat. That is the trap most content sites fall into. I rebuilt one calculator around the real search intent behind its impressions. I am judging the strategy on whether that page moves. If it does not, I will say so.
A 3D printing design studio I have run since 2019. I design original parametric models in Fusion 360 and Blender, through a 2D vector to 3D pipeline. The lines include geometric state and zodiac magnets, Celtic knotwork, dive coasters, Dungeons and Dragons coasters, jewelry, and functional organizers. I registered it as an LLC with its own storefront. It has fulfilled orders, retail display fixtures, and commissioned client work.
The published catalog is the part anyone can check. 15 models on MakerWorld, with roughly 1,400 downloads and 2,100 prints. That is 2,100 times someone took a design of mine and ran it on their own machine. The Earring Display Stand alone has been printed over 500 times.
A native iOS app that put live flight status and destination weather in one place. Schedules, delays, cancellations, and gate changes sat next to current conditions and a five day forecast. Push alerts fired when either one changed.
This was my first full product. I wrote the white paper, the competitive analysis, the tiered pricing model, and the feature roadmap, then built and shipped it. Most of what I know about scoping a product I learned from the parts of this one I got wrong.
A grid based 3D base builder for iPhone. You run an orbital debris cleanup business. Launch pads, barracks, and hangars are the navigation, and each building opens its own function. It plays like Fallout Shelter and SimCity, in orbit.
I adapted it from my father's book, Space Debris: Clearing the Final Frontier. He is my first TestFlight tester. That is a quality bar, and a reason to keep the release branch clean.
A pixel art iPhone tycoon. You start with one technician and a truck, then build a pest control company. You hire, upgrade the depot, and send crews out on jobs. It is in TestFlight beta.
This grew out of PestTech. Months inside the real industry gave me the domain detail a pest control game needs, and building the game clarified the operational loop the real product has to model.
A side scrolling shooter and a base builder, made for phones first. You upgrade rooms in a bunker, then raid sites to recover alien technology. It runs in the browser, with no install.
An AI generated desert art brand with an audience of roughly 11,000. The shop is open at thedigitalcacti.com, with prints, stickers, shirts, and color change mugs. A five stage pipeline generates, curates, produces, stages, and reports. Nothing publishes without an approval gate.
I grew the original account from zero to about 1,800 followers by hand, then lost it to an automation policy strike I caused myself. The rebuild keeps automation on production and takes it off engagement.
I used to run a self hosted assistant on an Obsidian vault. That setup is retired.
Now I use Grok Bot with a team of specialist bots, one per project. Code they write goes through a separate review step before it merges.
I went from a machine shop to battery research, then pharma marketing, and now enterprise people data and AI systems. Sixteen years. I build the whole measurement system, and I have stood on a factory floor where a wrong number on a dashboard had consequences. I use that same instinct when I evaluate a model.
I build, and I measure. I would rather ship a rough working thing and judge it honestly than present a polished plan for something that never gets built.
Claude API and the Agent SDK in production. Retrieval augmented generation over vector databases, with chunking, reranking, and agent orchestration. Local inference for anything that should not leave my machine.
Golden question sets, recall and precision at k, citation and groundedness scoring, lift over a random baseline, and refusal behavior. Runs are versioned, so a change in the system shows up as a change in the numbers.
Next.js and React on the front. FastAPI and Python on the back. Postgres and MongoDB for storage. Deploys go to Vercel and Render. React Native and Expo for iOS.
SQL, Python, BigQuery, and GA4. Tableau and Looker Studio for delivery. I can own a question from raw event data through to the decision someone makes on it.
Primary data over vendor summaries. Named targets with dates. An unevaluated model is a demo. Say plainly when something failed. Impressions, pageviews, and dashboards are not outcomes.