Real systems, labelled honestly.

ProblemA consultancy had two decades of methodology fully written down — and still answered every client question by hand, digging through a shared drive and re-explaining the context to a chatbot each time. In their words: the intelligence is all there, it just isn't connected.

BuildA private assistant their consultants can ask in plain language. It answers from the firm's own methodology and project files, shows which document each answer came from, and leaves the judgement call to the human — the constraint the client set on day one.

ProblemThe client believed their site was invisible to AI search and assistants, but had no diagnosis of why.

BuildA seven-page audit of AI citability, schema markup, and brand authority — finding a stale address in schema, missing org fields, and a misdated founding year, with a prioritised 30-day fix plan.

ProblemHigh-end enquiries arrived through a web form, sat in an inbox, and waited up to 48 hours for a manual reply — losing premium clients to faster competitors.

BuildAn n8n webhook that lets Claude read each enquiry, score it against budget and event signals, then route it down one of four paths: book, redirect, ask for missing info, or escalate a rush event.

ProblemAn executive needed the morning's incoming reports condensed before reaching the office. The material arrived as dozens of separate emails, and reading them on a commute was not an option.

BuildA scheduled workflow that collects the day's reports, has a model draft a bilingual summary, then runs a second pass that checks that draft against the source material and returns a patch list. Only the corrected version is sent — into an inbox, so a phone can read it aloud.

ProblemI wanted a portfolio piece that was also a real, working product — not a screenshot.

BuildA logged-in user submits a lead; a Trigger.dev task calls Claude and streams the structured qualification — tier, score, reasoning, next move — back to the browser in real time. Gated with Supabase auth, metered with a daily quota.

ProblemCreators waste hours reformatting one idea for every platform — and one failed post can break a whole publish run.

BuildA decoupled n8n system: a message in Telegram is rewritten by AI, approved, then published to LinkedIn, Threads, and Facebook in parallel. Each platform is fault-isolated, with a summary report if one fails.

ProblemManual prospecting on LinkedIn is slow and easy to get wrong.

BuildSearch by keyword, scrape profiles via Apify, let AI score each lead 1–10, and save the qualified ones to Google Sheets with a ready-to-send, personalised opener.

ProblemSaving new words is easy; turning them into something you actually review is not.

BuildAn n8n template that wires Telegram to OpenAI and Notion: send a word by text, voice, or photo and the bot returns a structured card — definition, translation, part of speech, example — then files it in Notion.