Full-Stack TypeScript Engineer (React / Next.js)

We run four TypeScript products — Next.js, Fastify, TanStack Start and Express — and we need a second engineer who can pick up any of them and ship. On-site in Islamabad / Rawalpindi. If your last four years were a single framework, read the stack section before you apply.

Why this job exists

AgentZap (DialogX IO) runs four products in parallel. All TypeScript, all at different stages:

• AgentZap — AI phone receptionists for small and mid-sized businesses. Agent configuration, integrations, post-call workflows, billing. Live, with paying customers.
• Gracero — all-in-one SaaS for med spas and cash-pay wellness clinics. Scheduling, POS, charting, programs. Early build.
• HysabOne — accounting and business software, desktop and web. Offline-first; it syncs when it reconnects. Rewrite in progress.
• Our connector platform — integrations with vendors who have no public API, where we reconstruct the interface ourselves. Live, with four connectors in production.

Today one person moves fluidly between all four. That is the bottleneck this job removes. You are not being hired to maintain one app. You are being hired to be the second person who can pick up any of the four and ship.

What you will own

Outcomes, not a task list.

• Shipped product surface. You take a ticket with acceptance criteria and return working, deployed, tested software. There is no QA team behind you. “It works on my machine” is not a status.
• Full vertical slices. Database migration, service layer, API, UI, deployed. You do not hand off at a layer boundary, because there is no one on the other side.
• Third-party integrations that survive contact with reality. OAuth flows, webhooks, private APIs, vendor sessions that expire at 3am. Roughly half this job is making someone else’s system behave.
• Your own verification. You test against the real thing before you call it done. Passing mock tests proves the code is shaped right, not that the vendor accepts it.
• Context that outlives you. Every repo carries its own agent instructions. When you learn something non-obvious, it goes into the repo — not into your head, and not into a DM.

The stack The shared spine — every day, on all four products

• TypeScript, strict mode. No any.
• React 19
• Tailwind with Radix / shadcn
• Postgres with a real ORM (Drizzle on two products, Prisma on one)
• Zod at every boundary
• Vitest and React Testing Library
• pnpm workspaces and monorepos
• Git and GitLab

Where the products diverge — we expect you to learn the ones you don’t know

• AgentZap: Next.js 15 (App Router), tRPC v11, Drizzle, BullMQ and Redis, Stripe, Twilio, Retell, Sentry
• Gracero: Fastify, Prisma, React with react-router, Vite, background workers, a CI-enforced package boundary
• HysabOne: TanStack Start and TanStack Router, Tauri for desktop, PowerSync for offline-first sync, Drizzle
• Integration-as-a-service Platform: Express, Playwright, the MCP SDK, libsql / Turso, node-cron, Ably

Be honest with yourself about this next line. We are not asking you to already know Tauri, PowerSync and MCP. We are asking whether you are the kind of engineer who has repeatedly learned an unfamiliar framework well enough to ship production code in it within a couple of weeks. If your last four years were one framework, this role will hurt.

How we work with AI — read this part twice

This is not a company that “allows” AI tools. Our repos are built for them: every project carries agent instructions, we write our own MCP servers, we author reusable agent skills, and we run several coding agents in parallel across git worktrees.

We expect you to produce the output of a small team. We also expect you to be more rigorous than a team, because generated code fails in quieter ways than hand-written code does.

What good looks like here

• You drive an agent through a real feature, schema to deployed UI, and the diff looks like it belongs in the codebase — because you gave it the codebase’s conventions up front.
• You read every line before it lands. You can explain any part of your pull request without opening it.
• You notice when the agent invented an API that does not exist, silently widened a type, wrote a test that asserts nothing, or “fixed” a failing test by deleting the assertion.
• You improve the tooling itself: sharpening the agent instructions, writing a skill, building an MCP server so the next task is cheaper for everyone.
• You know which tasks to hand to an agent and which to type yourself — and you can say why.

What gets you rejected

• Volume without verification. Large pull requests you cannot defend line by line.
• “The AI wrote it” offered as an explanation for a bug. It is your code the moment you open the pull request.
• Claiming done on offline evidence when a live check was available.
• Never having used these tools at all. This is a genuine requirement, not a nice-to-have.

Requirements

All of these. If you do not have them, please do not apply.

• 3+ years of professional TypeScript in production — real users, real incidents.
• Shipped and maintained a React application that real people used. Not a course project, not a clone, not a repo with one commit.
• Built server-side TypeScript — REST, tRPC or GraphQL — including the database layer.
• Written your own database migrations with an ORM, confidently.
• Implemented at least one OAuth2 or webhook integration end to end, including the parts that break: token refresh, retries, replay, expiry.
• You write tests unprompted. Your pull requests arrive with them.
• Daily, serious use of an AI coding agent (Claude Code, Cursor, Copilot, Codex or similar) on production code — with an opinion about where it fails.
• Strong written English. We work async and in writing.
• Able to work on-site in Islamabad / Rawalpindi, five days a week, long term.

Nice to have

These make you stand out. They are not required.

• Multi-tenant SaaS: tenant isolation, per-tenant data scoping, audit trails
• Monorepo experience, especially enforced module boundaries
• Offline-first or sync-engine work (PowerSync, ElectricSQL, CRDTs, Replicache)
• Desktop app packaging (Tauri, Electron)
• Browser automation at production quality (Playwright), or working with a private closed API
• Background jobs and queues (BullMQ, Redis)
• Stripe billing, especially usage-based
• Having written an MCP server or a reusable agent skill
• Voice AI product experience (Retell, Vapi, LiveKit, ElevenLabs)

How to apply

Your application needs three things. Applications without them are not read.

• A link to something you shipped that real people use. A live URL is preferred; a repo you can walk us through is acceptable. Tell us in one line what you built versus what the team built.
• One paragraph on an AI-assisted change you shipped — what you handed to the agent, what you kept for yourself, and one specific thing it got wrong that you caught.
• Your CV, including your home town.

Location

This is an on-site role in Islamabad / Rawalpindi, five days a week. Both twin cities work equally well for us. We are hiring for the long term and we do not offer relocation, so please apply only if the twin cities are somewhere you are settled rather than somewhere you are posted. Tell us your home town in your application — we ask everyone, and it is part of how we shortlist.

Job type: Full-time, on-site.

Pay: Rs100,000.00 - Rs185,000.00 per month

Work Location: In person