RalphJavenBeso

Problem

An Australian logistics company couldn't fit any off-the-shelf project management tool to their workflow — 12 distinct user roles, three project types, and a hierarchical company → segment → brand scoping model that no SaaS supported.

Approach

Designed a Postgres schema across ~18 entities, then built a React + Express + Prisma application from scratch. Implemented role-based access control with permission flags, Gantt scheduling with task dependencies, time logging, comments with @mentions, and a notification system. Used AI tooling as a build multiplier, owning every architectural decision and reviewing every line.

Outcome

Shipped a production-ready replacement for the planned ClickUp rollout in ~390 hours over 26 days. 92 UAT cases documented. Deployed to production on Vercel + Railway with separate staging environments.

Problem

Manually scrolling through LinkedIn, WeWorkRemotely, and RemoteOK is hours of work to surface a few relevant roles — and most listings are a 30-second skim before moving on.

Approach

Building an n8n workflow that pulls live job listings from RSS feeds every 6 hours, scores each role 1–10 against my CV via Claude API, drafts a personalised cover letter for the top matches, and delivers a daily digest to email and Slack with everything stored in Notion.

Outcome

Turns 2+ hours/day of job-hunting noise into a 5-minute morning review of pre-vetted, highest-fit roles. Replicable for any candidate with a CV and target keywords.

Problem

Action items, decisions, and key topics buried in 60-minute meeting recordings — turning recordings into useful, distributed outputs is a manual job nobody wants to do.

Approach

Building an n8n pipeline triggered on file upload to a Drive folder. Whisper API transcribes audio, Claude API extracts and categorises action items, decisions, and topics, then branches the output: tasks to Notion, decisions to a log database, and a summary email to attendees.

Outcome

Multi-step LLM pipeline that converts a meeting recording into structured, distributed output in under 90 seconds. Generalisable across teams of any size.

Problem

Most developer portfolios are template-shaped or undersell the work. I wanted one that reads like the way I actually think — clean, opinionated, and honest about how I use AI.

Approach

Built with Next.js 15, TypeScript, Tailwind, and Framer Motion. Custom theme system with light/dark, scroll-driven animations, marquee tech stack, and a process section that tells the truth about my workflow.

Outcome

A site that doubles as both a portfolio and a working code sample.