Description
Summary What: We are rebuilding our pharmaceutical commercial/regulatory ERP. The business logic is already fully specified - module by module, down to the field and gate level - in a complete written specification. We are not asking you to design the system. We are asking you to build it, and to build it primarily by piloting AI coding tools, not by hand-authoring every line. This is a defined-scope contract build (one core system, ~8 interconnected modules), followed by a lighter ongoing support arrangement. Remote, anywhere, with a few hours of daily overlap with GMT+8 required. During the build it is a full-time roll – do not apply if you are not available full time in the foreseeable future. Thereafter long-term part-time support by negotiation is ideal for us. We prefer a fixed cost deliverable and hourly backend but open to all proposals. The role Your job is judgement, steering, integration, and verification - not typing. We expect the AI to author the bulk of the code (PL/SQL business logic, the API layer, the front end) while you pilot it: directing it, reviewing what it produces, catching the subtle errors it makes confidently, integrating the modules so the seams actually hold, running it against real data, and proving it correct. If your instinct is to hand-craft everything over many months, you are the wrong person for this and will be unhappy here. If you already build this way and it excites you, read on. The stack • Oracle Database + PL/SQL — ALL business logic, gates, calculations, approvals, and security enforcement live here. This is an architectural law, not a preference. • ORDS — a thin API layer exposing the database logic. The API never reimplements logic; it exposes it. • Modern web front end (React or similar) — renders and collects input only. Never decides, never enforces, never orchestrates. Desktop only — no mobile, no responsive work required. The boundary is strict: the front end is a face; the database is the brain; the API is the wire between them. We will hold you to this. Required skills • Strong Oracle PL/SQL — you can read, write, and debug packages, triggers, and complex stored logic, and steer an AI agent through Oracle's quirks where it is weaker than in mainstream stacks. • Competent modern front-end (React or equivalent) and REST/ORDS API integration. • Genuine fluency with agentic AI coding tools (e.g. Claude Code or equivalent) — not 'I've tried ChatGPT', but a working practice of letting an AI, author and iterate in a real environment loop while you supervise, correct, and integrate. • Data modelling — normalizing a legacy structure into a clean relational model without losing business meaning. • Disciplined testing and integration instinct — you know that 'the demo works' is not 'the system is correct', especially for financial calculations. What you must agree to • The architectural law: all business logic in the database; the front end only renders and collects. No logic creeps into the front end for convenience. • AI-piloted authorship as the working method — you are comfortable being the pilot and reviewer, with the AI as primary author, and you are transparent about what was AI-generated. • Every gate override and every exception is logged with reason — the system's accountability backbone is non-negotiable. • You implement to the specification; business-logic decisions remain with us. Where the spec is ambiguous, you ask — you do not invent. • Source control, documented commits, and a reproducible build/deploy from day one. What you must prove before being hired A paid or unpaid trial slice. We will give you the complete specification for ONE module (with its real tables, logic, API, and desktop front end) and ask you to build it as a working vertical slice against a dev Oracle instance — using your AI tooling, in front of us or with a recorded/inspectable process. We are evaluating four things: • Do you reach for AI tooling naturally and pilot it well — or do you fight it and hand-type? (We want the former.) • Does your code respect the database-is-the-brain boundary? • When the AI produces something subtly wrong, do you catch it? • Can you make the numbers come out correct — verifiable against our existing tested model? Speed matters and we expect this to be fast — but a correct, well-bounded slice beats a fast, leaky one. The trial is how we both find out if this is a fit. How we work • We have done the architecture. You are not starting from a blank page — you are implementing a detailed, locked specification, which is exactly the situation in which AI-piloted building is fastest and safest. • Trust verification is fast on our side: we validate the ERP's outputs against an extensively-tested existing model. You will get quick, concrete right/wrong feedback on computed correctness. • Clear role separation: we own business logic and final authority; you own implementation, testing, deployment, and support. To apply: tell us how you currently use AI coding tools in real projects, what you have shipped this way, and your comfort with Oracle PL/SQL specifically.