Jam Susana AI Automation Engineer_

The client had a working GoHighLevel account, but every step of the customer journey needed someone to do it by hand. Leads were not being followed up, messages after sales calls were hit and miss, and customers who left were never contacted again.

I built seven automated stages, from the first sign-up to winning back customers who have gone quiet. What each person receives depends on what they do: whether they reply, how they score as a lead, and how their sales call went.

Finding technical talent meant searching patent filings, Google Scholar and LinkedIn by hand. It took hours per candidate, nobody scored them the same way, and it could not grow without a bigger team.

I built a launcher where the team chooses how to search, how strict the scoring should be, and which of 37 countries to cover. One click starts the right pipeline, and the results come back scored.

Each document meant pulling information from several places by hand and then writing it up. It took 2 to 4 hours a time, and the results varied from one writer to the next.

I built a pipeline that finds the relevant passages in the team’s own files, adds the latest figures from the database, and writes a structured draft for a person to check. It runs in n8n, with Supabase for the search and BigQuery for the live data.

New members were welcomed by hand, and not always the same way. The weekly Random Coffee pairings had nobody following up afterwards.

I built two bots. The first welcomes each new member in a conversation and checks that their profile is complete. The second pairs members every week, sends the introductions, and follows up on its own.

Old and past events kept appearing in the digests, and details that were missing when an event was first saved never got filled in later.

I rebuilt the date filter so it has both a start and an end, and changed how records are saved, so each one is updated when better information arrives instead of staying as it was first written.

Funding research was done by hand. Nothing was scored the same way twice, there was no order of priority, and no way to track where each program stood.

I built a pipeline that takes in each opportunity, scores it on more than 12 checks, and produces a ranked shortlist covering programs up to £5M.

There was no single view of how target communities performed on each platform, so partnerships could not be prioritised on real engagement numbers.

I built a Google Sheets dashboard that brings together follower counts, engagement rates and each community’s strongest platform, across 22 communities.

I had no organised way to track incoming leads, know what each one needed before a call, or move projects along without spreadsheets and memory.

I added a 6-step scoping form to my site. When someone submits it, the lead appears straight away on a private board, with the full brief, a booking link filled in with their details, and a contract template ready to send.

I wanted to show how candidate scoring can work as a product: a clear breakdown, a verdict, and a list you can sort, in a demo anyone can click through.

I built it to run entirely in the browser. The same name and field always give the same score, across four measures, and the candidate list is filled with made-up sample people.

Home-service businesses pay for every ad lead, then lose many of them by answering too late. I wanted to show the whole fix in one product anyone can click through: instant replies, follow-ups that stop on their own, and one place for a person to take over.

I built it to run entirely in the browser on sample leads from six trades. New leads arrive while you watch and some customers text back. Each one gets a confirmation in seconds, then a follow-up sequence that pauses the moment they reply and lands in a shared inbox. There is a 6-stage board with drag and drop, a leads table with CSV export, a team goal, reports by source, and a command palette.

I started from a template where every chart showed made-up numbers. There was no real data, the exports did not work, and the “live” updates were faked.

I rebuilt it on real on-chain data from The Graph and CoinGecko, with a demo mode so it still runs without an API key. I moved the code to TypeScript and added a loss calculator, a wallet lookup, pool comparison, a watchlist, working CSV export, light and dark themes, and automated tests.