Senior Full-Stack Dev (Next.js + Node/Prisma) — On-Demand for an AI-Built Legal SaaS

Read this whole post before applying. The first line of your proposal must be the word THEMIS-RTL followed by one sentence: the single hardest technical part of what's described below, and why. Proposals that skip this are deleted unread. Individual freelancers only. Agencies, dev shops, and "teams" are automatically rejected — do not apply. What this is I'm building an Arabic-first legal-memo SaaS (RTL, for UAE/Saudi law firms). Lawyers draft Arabic legal memos in a Word-like editor, route them for approval with comments and tracked changes, and export filing-ready PDFs (memo + exhibits merged, with letterhead separators). I am a non-developer founder, and I do most of the coding myself using AI tools (Claude Code). I am not hiring someone to build the whole thing. I'm hiring a senior developer to work alongside me in two modes: Unblocker. When I hit something the AI can't get right — a gnarly bug, a hard architecture decision, something subtle that's silently broken — I hand you that specific problem and you fix it properly. Executor. On days I can't code, I assign you scoped tasks to push the build forward, and you complete them to spec. So I need someone who is genuinely senior — strong enough to land in an unfamiliar codebase, diagnose what's wrong, and fix it cleanly — and who is also fluent at leveraging AI and open-source code (find a solid library, customize it to our needs) rather than reinventing everything by hand. Speed and judgment matter more than writing every line from scratch. The stack (must be strong in all of it) Frontend: Next.js (App Router), TypeScript, Tailwind, shadcn/ui — deployed on Vercel. Backend: Node.js, Express, Prisma, PostgreSQL. REST API, JWT auth, every record scoped by firmId. Storage / infra: S3-compatible object storage (Cloudflare R2), Redis + BullMQ for background jobs. The hard parts (this is what I'm really paying for): A rich-text editor that is RTL and Arabic-first — A4 pages, caption tables, nested lists, inline citations. Likely ProseMirror/TipTap. RTL done correctly is non-trivial; most people fake it with dir="rtl" and it falls apart. Tracked changes / suggestion mode — reviewers' edits become accept/reject suggestions tied to the live document, plus margin comments. This is the single hardest feature. A server-side PDF pipeline: render memo → PDF, merge exhibits with auto-generated separator pages, extract a single exhibit, compress to a per-file size cap (image downsampling), and split only at exhibit boundaries when a file exceeds the cap. Tooling likely from pdf-lib / pdfcpu / qpdf / Ghostscript. Multi-tenant data isolation — firmId enforced centrally (one Prisma/middleware layer), never hand-written per query. A user in Firm A must never reach any Firm B record. If half of that list is unfamiliar to you, this isn't your job — and that's fine. How we'll actually work You'll get scoped, written tasks (often a problem statement + the relevant files + what "done" looks like). You ship a fix or a feature, explain what you changed and why, and flag anything you think I got wrong. I want you to push back. If you see a better approach than what I propose, say so — with reasoning. I value a developer who challenges a bad idea over one who quietly builds it. Comfort with AI-assisted development is required. You should be the kind of dev who reaches for the right open-source library, reads the actual docs, and adapts proven code — not someone who hand-rolls everything or, at the other extreme, blindly pastes AI output without understanding it. Engagement Hourly, on-demand. No guaranteed minimum hours. Some weeks may be a few hours of unblocking; others, a couple of focused days of assigned tasks. Some overlap with Gulf Standard Time (Dubai, GMT+4) working hours for live handoffs is needed; you don't have to be in the region. Long-term potential: if this works, the relationship continues well past the MVP. To apply Start your proposal with THEMIS-RTL + the one-sentence "hardest part" answer (see top of this post). In 3–5 sentences, tell me about one time you dropped into an unfamiliar codebase and fixed a hard bug or made a hard call — what was broken, how you figured it out, what you did. Link me to specific work you've done with: a rich-text editor (ProseMirror/TipTap/Slate/Quill), RTL/Arabic UI, or server-side PDF manipulation. Real repos, PRs, or live products — not a generic portfolio site. One line on how you use AI tools in your day-to-day development. Shortlisted candidates get a short written technical quiz, followed by a 15-minute call. Do not send a generic proposal