Yuelin Liu · Melbourne · Full-Stack Developer · AI & LLM Applications

I build reliable full-stack software, with a focus on AI/LLM applications and developer tools.

I work across product UI, APIs, authentication, data workflows and model-integrated features—turning messy requirements into reliable software.

Download ResumeContact
LinkedInGitHubWriting

Selected Work

DSH Conversation Exporter product screenshot

DSH Conversation Exporter

Exports full DeepSeek Harness conversations or selected whole turns as clean Markdown for reading, Git and cross-assistant handoff.

Built and packaged as a DSH Web plugin · 2026

Preserves Markdown and Unicode, keeps selective turn handling bounded, and excludes reasoning, tool activity and runtime metadata from exports.

Stack · JavaScript · Node.js · DSH plugin APIs · Markdown

Read case studySource
MoneyGuard product screenshot

MoneyGuard

Turns a timecard photo into a weekly wage, spending and surplus audit.

Designed and built the end-to-end pipeline · 2026

Keeps ledger calculations local, validates model-generated OCR with Zod, and streams audit prose through separate providers.

Stack · TypeScript · Node.js · Gemini LLM · DeepSeek LLM

Read case studySource

More Engineering Work

  • AI HarnessCurrent work

    GitHub-native AI-assisted engineering workflow built around NEXT → IMPLEMENT → ACCEPT → SHIP, exact PR-head SHA acceptance, deterministic verification and explicit human approval.

  • Melbourne University Ultimate Club Platform

    Revisited a Next.js/MongoDB application to centralise server-side authorisation and turn authentication and write-path defects into regression tests and CI.

  • Client Data Delivery Validator

    Python/pandas validation workflow for synthetic client-delivery CSV data with structured validation results and 12 pytest regression cases.

Experience

Hands-on development across prospective-client prototypes and an early-stage AI SaaS codebase, focused on working implementations, API integration and diagnosing inherited workflows.

Independent Developer

Prospective Client Demos for a Melbourne Consultancy

Sep 2025 - Present

Melbourne, Australia

  • Build demonstration applications for prospective clients, including a specialty coffee retailer and a drone training provider, translating informal briefs into working prototypes for business pitches.
  • Implement authenticated, database-backed prototypes with Next.js 15 App Router, Clerk OAuth and Convex as requirements become clearer.

Next.js 15 App Router · Clerk OAuth · Convex

Full-Stack Developer

ByteCroniX — Early-Stage AI SaaS

Mar 2025 - Jun 2025

Melbourne, Australia

  • Implemented OpenAI function-calling flows in a Node.js/Express codebase for AI-assisted product interactions.
  • Added timing instrumentation to an inherited points-summary path and identified current/proposed scoring services as the dominant latency contributors.
  • Profiled an inherited points-summary pipeline and parallelised independent service calls, reducing an observed local integration-test response from 26.7s to 5.6s.

Node.js · Express · OpenAI Function Calling · Docker/Postman Integration Testing

How I Build

I make the workflow concrete, measure what is actually happening, and turn fragile assumptions into explicit checks.

I start with the workflow, not the feature.

I turn an unclear problem into a concrete working flow before adding abstraction, polish or extra infrastructure.

  • Built prospective-client prototypes for a specialty coffee retailer and a drone training provider from informal business briefs.
  • Built MoneyGuard around a real wage-checking workflow, separating model-assisted extraction from deterministic calculations.

I measure before I optimise.

I instrument inherited behaviour, trace where time or state is actually going, and change the smallest layer that explains the problem.

  • Instrumented ByteCroniX's inherited points-summary path, identified scoring services as the dominant latency contributors, then parallelised independent calls.
  • Revisited Melbourne Ultimate, traced silent session-role failures to fragmented authentication usage, and centralised session reads behind one database-verified guard.

I turn assumptions into enforceable boundaries.

I prefer schemas, explicit write rules and regression checks over relying on UI conventions, model behaviour or code-review memory.

  • Schema-validated MoneyGuard OCR output with Zod before deterministic wage calculations consume it.
  • Turned Melbourne Ultimate authorization and mass-assignment findings into automated regression tests and CI checks.
Product UI
React 19 · Next.js 15/16 · TypeScript · Tailwind CSS
APIs & Data
Node.js · Express · REST APIs · MongoDB · Convex
Auth & Delivery
Clerk OAuth · NextAuth · GitHub Actions · Docker · Vitest
AI & Validation
OpenAI Function Calling · Gemini · DeepSeek · Zod

Contact

I am based in Melbourne and open to full-stack software engineer opportunities. Email is usually the cleanest path for role details, interview scheduling and technical context.

Prepare a role note

This opens a pre-filled email draft to liuyuelintop@gmail.com. It is intentionally lightweight so recruiters and hiring managers can start a real conversation quickly.

Message focus *