3. Mandatory requirements

3.1 Experience

- Minimum 2 years of professional development experience (internships and academic

projects excluded).

- At least one web application and one mobile application shipped to real users — live URL, App Store, or Play Store listing required as evidence.

- Demonstrable experience working on a system with real load: concurrent users,

background jobs, or a database beyond a few thousand rows.

3.2 Web application development

- Frontend: React (or Vue/Angular) with TypeScript. Component architecture, state management, client-side routing, responsive layout. -

- Meta-framework: Next.js, Remix, or Nuxt — server-side rendering, static generation,

and route-level data loading.

- Styling: Tailwind CSS or an equivalent utility/design-token system.

3.3 Mobile application development

- Cross-platform: React Native (Expo) or Flutter. Native iOS/Android experience is an

acceptable substitute if paired with web skills. -

- Platform essentials: navigation, offline/local storage, push notifications, deep linking,

permissions handling.

-Release mechanics: build signing, store submission, review-rejection handling,

OTA/CodePush-style updates.

3.4 Backend & data

- Node.js (NestJS, Express, or Fastify) or Python (FastAPI/Django) or Go. - - -

- REST API design; GraphQL exposure is an advantage.

- Relational database work in PostgreSQL or MySQL: schema design, normalisation,

migrations, indexing, and reading a query plan.

- Redis or equivalent for caching and session/rate-limit storage.

- Authentication and authorisation: OAuth 2.0 / OIDC, JWT handling, refresh-token

rotation, role-based access control.

3.5 Scalable architecture

The candidate must be able to explain and implement, not just name:

- Stateless service design that permits horizontal scaling behind a load balancer.

- Caching strategy across layers (CDN, application, database) and cache invalidation.

- Asynchronous processing with queues and workers (SQS, RabbitMQ, BullMQ, Celery,

Kafka).

- Database scaling basics: connection pooling, read replicas, pagination that doesn't

degrade, N+1 query avoidance.

- When a modular monolith is the right answer and when to split out a service — and the

cost of splitting too early.

- Idempotency, retries, timeouts, and graceful degradation on third-party failures.

- API versioning and backward compatibility, which matters more on mobile because old

app versions stay in the wild.

3.6 Cloud, delivery & operations

- One major cloud provider at working depth: AWS, GCP, or Azure. - - - -

- Docker; container deployment via ECS, Cloud Run, Kubernetes, or a managed PaaS.

- CI/CD pipelines (GitHub Actions, GitLab CI) including automated test and build gates.

- Infrastructure as code — Terraform, Pulumi, or CDK.

- Observability: structured logging, error tracking (Sentry), metrics and alerting; ability to

diagnose a production incident from telemetry.

- Git at a team level: trunk-based or GitFlow, meaningful commits, code review

participation.

3.7 AI-assisted development capability

This is the differentiating requirement and must be assessed directly, not taken on trust. - - Daily working fluency with at least two of: Claude Code, Cursor, GitHub Copilot,

Windsurf, Codex, or a comparable agentic coding tool.

- Specification skill: can decompose a feature into small, independently verifiable units

of work and write the context, constraints, and acceptance criteria an AI agent needs to

get it right on the first or second attempt.

- Context management: maintains project-level instruction files (AGENTS.md /

CLAUDE.md / rules files), curates what the model sees, and keeps the codebase legible

so generated code stays consistent with existing patterns.

- Critical review of generated output: reads every line before it merges. Specifically

checks for hallucinated APIs, non-existent packages, missing error handling,

unparameterised queries, exposed secrets, unsafe defaults, and quietly quadratic

algorithms.

- Knows when to stop prompting. Recognises the point at which re-prompting is slower

than writing the code manually, and switches.

- Test-first discipline with agents: uses tests and type checks as the verification loop

rather than relying on visual inspection of AI output.

- Dependency and licence hygiene: does not accept AI-suggested packages without

checking maintenance status, install size, and licence.

- Debugging without assistance: can find and fix a bug in unfamiliar code using a

debugger, logs, and reasoning, with AI tools switched off.

3.8 Non-technical requirements

- Written communication strong enough to specify work clearly — this correlates directly

with AI-assisted output quality. - -

- Comfortable with ambiguity and short feedback cycles; asks before building the wrong

thing.

- Takes ownership of defects reaching production; no deflection onto the tooling.

- Professional English, spoken and written.

4. Desirable (not disqualifying) 

- Real-time features: WebSockets, SSE, presence, collaborative editing.

- Payments (Stripe, PayPal, local gateways) and PCI-adjacent handling.

- Multi-tenant SaaS data isolation patterns.

- Analytics and product instrumentation; A/B testing infrastructure.

- Accessibility to WCAG 2.1 AA.

- Performance work: Core Web Vitals, bundle budgets, mobile cold-start time.

- Open-source contributions or a public portfolio of AI-assisted builds.

- Exposure to LLM feature work inside products (RAG, embeddings, evals) as distinct

from AI-assisted coding.