I build systemsthat moveteams forward.

01/Systems & Platforms

PeopleOS

HR, attendance and statutory payroll for ~300 people

The problem

HR ran on a 291-column master workbook, a legacy attendance app, paper leave forms and a macro-driven payroll spreadsheet. Nobody could answer who reports to whom, on what contract, since when — and no month's payroll could be proved to have come from a specific set of rules.

What I built

A full HR platform: 17 domain modules over a 102-model Postgres schema, an append-only attendance ledger, and separately versioned rule engines that price statutory payroll in exact decimal arithmetic.

My role

Sole engineer and de-facto product owner. Domain model, rule interpretation with Accounting and HR, release process, and production ownership.

What changed

Replaced spreadsheets and paper with one auditable system of record, live on managed cloud infrastructure since the August 2026 cutover.

• TypeScript
• Next.js 16
• React 19
• PostgreSQL
• Prisma 6
• decimal.js
• Zod 4
• AES-256-GCM
• argon2
• React-PDF

02/AI & Automation

Ads Review Automation

A deterministic rules engine with an LLM allowed only to explain it

The problem

The weekly paid-media review was a manual pass through Ads Manager: read ROAS, CPA, CPM and CPV per ad, decide what to pause, re-type the conclusions into a document. The thresholds lived in someone's head, so the same numbers produced different decisions on different weeks.

What I built

A staged Python pipeline that fetches ad performance, scores every ad against thresholds held in one JSON file, checks that pausing an ad will not leave a product with zero live coverage, and ships the review in four formats.

My role

Sole designer and implementer. Wrote the design specs, the rules engine, four prompt versions, the exporters, and merged the work through reviewed feature branches.

What changed

Turned a remembered judgement into a reproducible one: the thresholds that drive the decision now feed the engine, the spreadsheet templates and the LLM prompt from a single file.

• Python 3
• pandas
• Meta Marketing API v25
• OpenAI API
• gpt-4o-mini
• gspread
• google-api-python-client
• Playwright
• markdown-it-py
• openpyxl

03/Decision Systems

Slow Seller Decision Engine

A recommendation per style, and a spreadsheet it cannot damage

The problem

Merchandisers decided each season what to keep, sell out, mark down or drop — but the inputs were assembled by hand and by a brittle Apps Script. Four sales columns carried end dates from four different days, status grouping was a hard-coded list that broke whenever someone typed a new wording, and two engine inputs were placeholders rather than real data.

What I built

A Next.js app that runs a seven-stage sync — read the sheet, load Shopify products and orders, match style and colour, aggregate two rolling windows, run the engine, write back — streaming progress over Server-Sent Events.

My role

Sole developer. Both integrations, the decision engine, the test suite, and a 1,032-line README written for the operator as much as the developer.

What changed

The same decision, now sourced from live Shopify data, reproducible, and structurally unable to overwrite a formula, a header, or a colleague's edit.

• TypeScript
• Next.js 15
• React 19
• Tailwind CSS 3
• Radix UI
• Zod
• Shopify Admin GraphQL
• Google Sheets API v4
• Server-Sent Events
• Vitest

04/Internal Tools

Order Export for Production Planning

Removing the developer from a routine data request

The problem

Production planners needed one row per line item, each tagged with the warehouse that will actually ship it. Shopify's native export does not produce that shape, and Shopify exposes several non-equivalent 'location' values — so 'which warehouse' is a decision problem, not a field lookup.

What I built

A stateless FastAPI service where the planning team picks a date range, previews the orders in-browser, and downloads Excel or CSV — behind company Google sign-in, with nothing customer-related written to disk.

My role

Sole developer: requirements analysis, Shopify schema investigation, architecture, implementation, tests, containerisation and the cloud deployment.

What changed

A non-technical team now self-serves data that previously required a developer to run local scripts on request.

• Python 3.12
• FastAPI
• Pydantic v2
• httpx
• tenacity
• Jinja2
• XlsxWriter
• gspread
• Docker
• Cloud Run

05/AI & Automation

Seasonal Catalog Media Pipeline

~1,000 images, 165 products, and no click-by-click afternoon

The problem

A seasonal drop meant getting roughly 1,000 images into WordPress under a strict naming convention, then telling each of 165 products which images are its own, in which order, with which colour label. Source filenames come from a design export and do not match the catalog's colourway names — colours appear in a different order, get abbreviated, and style names contain typos, so a naive string match silently binds the wrong swatch to the wrong colourway.

What I built

Seven standalone Python CLI tools, standard library only, forming a staged pipeline from a design export to live collection pages — every writing tool dry-run by default.

My role

Sole developer and operator. Designed the naming convention, wrote the matching heuristics and the WordPress client, and ran the pipeline against production.

What changed

A drop that would otherwise be a click-by-click afternoon in the media library and ACF editor became a reviewable plan you approve and apply.

• Python 3.14
• Standard library only
• WordPress REST API
• Advanced Custom Fields
• Application Passwords
• CSV
• WebP / PNG

06/Developer Tooling

Agent Session Observability

Watching a dozen AI coding sessions without tabbing through windows

The problem

Running several AI coding sessions at once leaves no way to know which one is working, which is silently blocked on a permission prompt, which errored and which finished — short of tabbing through every terminal and editor window. There is no status API. The only signal is the session registry, the process table, and multi-megabyte JSONL transcripts on disk.

What I built

A zero-dependency Electron dashboard that derives full session state purely by observing files — an incremental transcript digester, an 11-rule state machine, per-model cost arithmetic, and a probabilistic ETA with a confidence band.

My role

Sole author and architect. Wrote the build contract that pinned the module shapes, then integrated the modules implemented against it.

What changed

Answers at a glance what previously required opening every window: what is each session doing, how far along, roughly how much longer, and what the token usage would have cost.

• Node.js 22 (ESM)
• Electron 34
• Server-Sent Events
• node:fs streams
• JSONL stream parsing
• Vanilla JavaScript
• Inline SVG
• Bash