AI Tools, Automations & Custom Systems

A resume and career-tooling platform that turns a rough work history into an ATS-ready CV, then keeps generating the surrounding material — cover letters, professional summaries, interview preparation — from that same profile rather than making the user start over each time.

Most applicants are filtered out by automated screening before a human reads a word. Rewriting a CV for every application is slow, so people send one generic version everywhere and never learn why it failed.

Node.js and Express behind a hosted LLM inference layer for generation, Stripe for subscriptions and one-off credits, and a shared component library so each new tool ships without redesigning the interface. Deployed on Railway.

An AI sales assistant embedded in a Shopify storefront. It answers product questions in the shopper's own words, narrows a large catalogue down to a short list, and moves the conversation toward checkout. Running on the Neurolab and P1 Peptides stores.

Supplement and peptide buyers arrive with specific, research-heavy questions. A static FAQ page does not answer them, and the questions usually land outside working hours — so the cart is abandoned long before anyone can reply.

Grounded on the store's live product catalogue rather than a general-purpose model, so answers stay tied to real products, stock and pricing instead of being invented. Embedded through the Shopify theme so it travels with the storefront, with conversation handling and product lookup behind an API layer.

An interactive configurator for an interior furnishing studio. A client picks property type, room count and specification level, and the tool assembles the matching package and prices it live on the page.

Every enquiry was quoted by hand. Prospects could not get a number without booking a sales conversation, so early-stage buyers dropped out — and the studio spent hours quoting jobs that were never going to convert.

A real pricing engine rather than a form that emails an enquiry. Tiers, per-room rates and rules are stored as editable data so the studio can change pricing without touching code, and the configurator recalculates against them instantly. WordPress with ACF as the data layer and a JavaScript front end.

A research pipeline that builds qualified prospect lists from public business data, verifies and deduplicates the contacts, and runs outreach in controlled batches — all driven from one dashboard.

Manual prospecting does not scale, and without a shared record the same businesses get contacted twice across different campaigns. Bad data wastes send volume and damages sender reputation.

Python collection services reading from public sources, including OpenStreetMap's Overpass API for mapped businesses, with a SQLite store behind a FastAPI layer. Three-stage deduplication means no contact is approached twice across campaigns, and every step runs from the dashboard rather than the command line.

Payment integrations and back-office automation — getting payment taken reliably, and getting the resulting orders into fulfilment without anyone retyping them.

Some sectors cannot use mainstream checkout providers, so payments have to be routed through specialist processors. And once an order lands it is usually re-entered by hand into a courier or warehouse system, which is slow and starts producing errors the moment volume rises.

Custom middleware sitting between the storefront and third-party systems: gateway integrations for merchants outside mainstream processing, plus a data pipeline that maps store orders into a logistics provider's required format, exports item and stock records, and pushes fulfilment status back to the store.