Selected work

Systems I have actually shipped

Six projects, described in enough detail that you can judge whether the thinking transfers to your problem — which is the only reason a case study is worth reading.

Financial servicesIn-house

Real-time eligibility alerts during live customer conversations

Manual checks removed from the sales floor

The problem

During a loan appraisal, sales staff could not see whether the customer qualified for a bigger product. The data existed, but only behind database queries they were not allowed to run. So they messaged the underwriting team and waited. By the time an answer came back, the customer had usually gone home.

The approach

I put a safe layer in front of the database so the check could run without exposing it. A Python service watched for appraisals, ran the check, and wrote the result to a controlled sheet. If the customer qualified, a WhatsApp message reached the loan manager while they were still in the room.

The outcome

The back-and-forth between sales and underwriting went away. The answer arrived during the conversation instead of after it. The system ran in production through business hours.

40%

Improvement in conversion

Stated on the resume for the High-LTV gold loan product, which this automation supported.

30%

Improvement in sales productivity

Stated on the resume for the same product line.

PythonGoogle Apps ScriptGoogle SheetsWhatsApp Business APIREST APIs

Built in-house as Program Manager at a fintech lender

Content operationsPersonal build

A research-to-draft content pipeline with no manual steps

Keyword research to publish-ready draft, unattended

The problem

Publishing regularly means repeating the same chain: find a topic, research it, structure it, draft it, load it into the site. Each step is easy. Together they cost hours. And it is the first thing to stop happening in a busy week.

The approach

I built a pipeline in n8n that finds trending keywords, hands the research to a language model, and prepares a finished draft straight into WordPress. Nothing moves between tools by hand.

The outcome

A draft arrives ready to edit rather than ready to assemble. It is the clearest example of what I build for clients: the expensive part of a job is rarely the thinking, it is the handoffs.

n8nAnthropic ClaudeREST APIsWordPress

Personal build — my own automation, running on my own stack

Insurance operationsIn-house

Automated detection and recovery of failed policy issuance

A failure queue that clears itself

The problem

A small share of insurance policies failed to issue. Finding them was someone's job. The operations team checked by hand, spotted the failures, and re-ran them one at a time. Dull work, easy to put off — and when it slipped, customers lost their policies.

The approach

I built a pipeline that spots failed cases on its own and re-runs them through the right APIs. Anything needing a judgement call goes to the operations team as an alert.

The outcome

Nobody watches the queue by hand any more. Failures are found and re-run automatically, so the queue clears whether or not someone remembers to look.

Google Apps ScriptREST APIsPostman

Built in-house at a fintech lender

Marketing operationsIn-house

Structured spreadsheet data turned into personalised outreach

Leads stop dying in a spreadsheet

The problem

Marketing leads piled up in a spreadsheet and sat there. Every hour a lead waits, the odds of a useful conversation drop. But contacting one meant a person opening the sheet, reading a row and writing a message.

The approach

A Google Apps Script that fires when the sheet updates, takes the row, and sends a personalised WhatsApp message the moment the lead arrives.

The outcome

Leads get a message straight away instead of waiting for someone to open the sheet.

Google Apps ScriptGoogle SheetsWhatsApp Business API

Built in-house at a fintech lender

Internal operationsPersonal build

A chat assistant that triggers internal workflows

Operational actions from a chat window

The problem

Routine internal jobs — check a status, start a process, pull up a record — each meant opening the right tool and knowing where to click. Small friction, repeated all day, and it all landed on whoever knew the systems best.

The approach

An n8n-powered Telegram assistant. You ask for something in chat and it runs the workflow behind it, in the place people are already typing.

The outcome

Routine requests get handled in chat instead of queueing up for the one person who knows where everything lives.

n8nTelegramLLM prompting

Personal build

CommunicationsPersonal build

Drafted email responses for repetitive correspondence

Routine replies drafted, not written

The problem

Most email is repetitive: notifications, follow-ups, and the same few replies rewritten slightly differently each time. It takes more attention than it is worth.

The approach

An email pipeline that handles notifications and follow-ups on set rules, plus an AI assistant that drafts the routine replies. Every draft waits for a person to approve it. Nothing sends unreviewed.

The outcome

The routine mail is written before you open it. What is left is reading and approving, not composing.

LLM promptingn8nGoogle Apps Script

Personal build

Ready to get the hours back?

A free 30-minute audit call. You leave with a written map of your automatable workflows and what each one is worth — whether we end up working together or not.

Email hello@lytheo.com Message on WhatsApp

Replies within 24 hours · No sales team — you talk to the builder