← Jobb

Technical Delivery Lead, AI Software

Budget: $15.0 - $25.0 HOURLY / FULL_TIME ⭐ 4.97 (183) Australia

python, typescript, postgresql, software-testing, technical

Föredragna kvalifikationer

  • Plats: Vietnam, Thailand, Malaysia, Philippines
  • Erfarenhet: Medel
  • Engelska: Konversationsnivå
  • Job Success: 90%+
  • Rising Talent föredras
  • Min. intäkt: $1,000+
Hi! We're looking for someone with 20-40 hours per week availability for a steady, reliable 5 star client. We build business systems for civil construction and contracting companies. You will take over the build phase of those systems from the Technical Director. You get a plan; you finish the software: built, tested, verified, handed over. Then you own it. You sit on the client calls, take their bugs and feature requests, decide what gets built and when, and keep them happy. How we work Almost no code here is hand-typed. It is produced by coding agents (Claude Code, Codex) under human direction. Your job is conductor and verifier first, typist second. The dangerous failure in this kind of work is "looked done, wasn't": the scheduled job that never ran, the deploy that reported success while the old code kept running, the send lane with zero deliveries ever. Every one of those gets caught by someone who demands evidence instead of reading a diff and believing it. That is the person we want. Spec first, in writing. Testing is most of the job: agents produce the code quickly, and the value you add is proving it works. Unit tests where they earn their keep, then the real path on dev and test, then a fresh adversarial review, with evidence kept at each step. Hand over in plain English the client's owner can follow. Async and self-managed: written specs and handovers, no daily standup. What you will do * Take a written plan and drive agents to build it: fresh branch, PR, CI, dev, then test * Test it properly: a contract of testable assertions before code, failing test first, boundary and permission cases, browser-level QA across viewports, then the real path exercised on dev and again on test * Verify with evidence: the number re-derived from the query, the job actually ran, the email actually sent, the agent verification loop run until it converges * Write the release note and handover a non-developer can test from * Run the client's bug and feature list: triage, prioritise, push back when needed, ship * Sit on client calls and turn what you hear into testable tickets * Work across more than one client without mixing their stacks, data or rules Who we are looking for * Evidence over belief. You do not trust a green CI tick, an agent's "done", or your own memory. You go and look. * Systems and adversarial thinking. You hold the whole flow in your head (UI, API, worker, DB, constraints) and you imagine the zero quantity, the NULL vs blank string, and the range widened in the UI but not the DB before anything breaks. * Conversational English * Judgement and restraint. You decide routine calls alone and escalate only the expensive ones. You challenge a spec with a reason, then commit. You leave alone what was not asked for. * You finish, and you communicate in plain English. Done means verified with evidence and handed over. You are as good on a call with a frustrated client as you are in a handover doc. * You own a backlog. A long list of bugs and requests does not scare you. You prioritise by what the client actually needs and you are comfortable saying "later" or "no". * You have driven coding agents seriously and been burned. You can name a specific thing your agent got wrong and what you changed in your process because of it. You will already have * Reads unfamiliar Python and TypeScript fast; can trace a number on screen back to its source query in under an hour * Postgres fluency: why AVG(rate) is wrong and SUM(a)/SUM(b) is right, window functions, NULL vs empty string on a unique constraint * Linux and shell (processes, logs, cron, env files, ownership) and a real Git mental model (worktrees, rebase, why a reused branch name is rejected) * Testing instinct: knows a test that cannot fail from one that can, writes the failing test first, and treats "tests pass" as the start of verification, not the end * Browser-level QA with Playwright or equivalent; knows a console error from a network error from a layout defect * Third-party API debugging: OAuth, rate limits, sandbox vs live, webhooks We will teach you Each client's stack, environments, deploy and rollback, runbooks and glossary. Our spec-first workflow and verification ladder. Construction and civil contracting vocabulary. Our house style and tooling. Industry experience is a tiebreaker, not a requirement.
Öppna på Upwork

AI proposal draft

Generate a short cover letter for this job. Edit before sending.

Sign in to generate an AI proposal draft.

Logga in