I build the tools,
not just the roadmap.
When AI is on every roadmap, the useful question isn't whether to use it, it's where. Building things myself is what lets me tell a real opportunity from a good demo.
My background is analysis, product, tech, data and delivery, not pure engineering. With AI I take things all the way. I integrate, test and ship. I use it for far more than discovery and user stories. I build tools, PoCs, integrations, sometimes full apps. That hands-on work is how I learn where AI actually belongs and how to build systems around it.
Shop inventory & automation
Syncs physical stock and prices into Shopify, recalculates cut-to-size product variants, flags anything that's gone negative, and emails suppliers a weekly restock plan. Python, Google Apps Script and Shopify, with secrets in Google Secret Manager and a full unit-test suite behind it.
Geolocalization analyzer & maps
Give it an address; it pulls land use, transport links and nearby competition from OpenStreetMap and Google Maps, routes the analysis across GPT-4o, Claude and Gemini, cross-checks their answers, and auto-generates a full PowerPoint, Word and interactive HTML deck — with sources cited.
Price & trend predictor — TCG / CCG
A daily price-history pipeline for a live trading-card-game market — Python, PostgreSQL, DuckDB, tens of millions of rows. A walk-forward backtest with a realistic spread model, an anti-bulk filter and zero look-ahead bias, so the signal you see is one you could actually have traded.
Price & trend predictor — Solana
Pulls wallet and transaction data from Solana through API integrations, cleans and aggregates raw on-chain activity, and folds scattered chat threads and signals into one clean feed with alerts — trend analytics a non-engineer can read and act on.
Customs & returns automation
Automating the customs and returns handling behind a shop — declarations, documents and the operational back-and-forth that usually eats people's time, not just the storefront.
TMS tooling evaluation
Integrated and ran Localizely, Crowdin, Localise and Tolgee end-to-end for a mobile app — tags, resources, namespaces, screenshots, projects — to judge fit hands-on rather than from a comparison table.
App-store reviews dashboard & analyzer
Custom dashboard pulling customer reports and reviews from Google Play, the App Store and social, with charts and automatic clustering of recurring problems — so the top issues surface instead of hiding in a feed.
Closed-Loop Digital Brain
The wiki holds one shared context and a loop runs on top of it. Customer signals and decisions go in. Agents work inside the rules and turn them into tickets, docs, QA hints, release notes, support FAQs and status digests. The wiki keeps itself current as it goes. The product assistant I'm building already lives on that loop, reading app-store reviews and drafting the release notes.
The thread under all of it is a product brain. One shared context, the wiki, with a loop on top. Signals go in, agents work inside the rules, product comes out, and the wiki keeps itself current.
The system is the easy part. The real question is when it's worth building at all. I'm building a product assistant on this loop. Full write-up in Thoughts, coming soon.