Senior Full Stack Developer

Senior Full Stack Developer AI-augmented applications · Next.js · TypeScript · Azure · LLM engineering Location: Remote Employment type: Full-time Level: Senior individual contributor (typically 5–8 years of professional experience) Role summary We are hiring a Senior Full Stack Developer to build AI-augmented enterprise web applications end to end — the interface, the services behind it, the retrieval and evaluation infrastructure that makes the model output trustworthy, and the deployment and observability that keep it running. This is a hands-on engineering role with an unusually wide surface area. You will work in TypeScript across a modern React and Next.js front end and a typed API layer, integrate large language models as production components rather than demos, and hold the result to enterprise standards for security, accessibility, and auditability. You are also expected to use AI development tooling as a normal part of how you work — across discovery, specification, implementation, testing, review, and documentation. Fluency with agentic coding tools is a required skill for this role, not a bonus. What you will build The applications are server-rendered React front ends over a typed API layer, with LLM-backed capabilities that must be grounded, measurable, and auditable rather than merely impressive. The recurring capability patterns you will own: • Structured extraction from free text. Turning an unstructured narrative into typed, validated fields — surfaced back to the user as editable values before anything downstream acts on them. • Grounded generation. Retrieval over enterprise content, with citations, source attribution, and an explicit confidence posture instead of unqualified assertion. • Multi-turn conversational flows. Persona-driven sessions with state carried across turns, scored against a rubric at the end. • Analytics and forecasting surfaces. Aggregation across operational systems, rendered as scores, trends, and forecasts that state their own uncertainty. • Agentic action against systems of record. The assistant invoking real workflows — ticket creation, escalation, scheduling — behind idempotent tools, explicit confirmation gates on irreversible actions, and a complete audit trail. • Embedded surfaces. The same capabilities rendered as panels and extensions inside third-party enterprise SaaS hosts and chat platforms, subject to their layout, security, and lifecycle constraints. • Accessible-by-construction interfaces. WCAG 2.2 AA as a build requirement: semantic structure, managed focus, keyboard parity, honored motion and contrast preferences, and a non-visual equivalent for every visualization. A representative baseline architecture: Next.js App Router with React Server Components and Server Actions, a tokenized design system on Tailwind with headless accessible primitives, authentication enforced at Edge Middleware, typed clients generated from an OpenAPI contract, model calls served by a separate AI service layer, and the whole thing deployed to Azure with structured tracing and cost instrumentation. Technology stack You will not touch every row on day one, and no candidate arrives with all of it. We are looking for demonstrated depth in the frontend and AI rows, working competence in the server, cloud, and quality rows, and the judgment to learn the rest quickly. Layer Technologies Frontend Next.js 14+ (App Router), React 18/19, TypeScript in strict mode, Tailwind CSS with tokenized design systems, Radix UI headless primitives, TanStack Query, Recharts or Visx for charting Server & API React Server Components, Server Actions, Route Handlers, Edge Middleware, Node.js 22, Python (FastAPI) for AI services, OpenAPI-first contracts with generated TypeScript clients and Zod validation AI & LLM Azure OpenAI, Anthropic Claude, and OpenAI APIs; retrieval-augmented generation over Azure AI Search or pgvector; structured outputs, function/tool calling, and Model Context Protocol (MCP); token streaming; prompt, context, and eval management Data PostgreSQL and Azure SQL, Redis, Cosmos DB, Azure Blob Storage; event-driven ingestion and medallion-style analytics layers Identity Microsoft Entra ID and Okta OIDC single sign-on, SCIM provisioning, role-based access control, tenant isolation, Auth.js Cloud & DevOps Azure App Service and Container Apps, Key Vault, Managed Identity, Application Insights, Front Door; Azure DevOps Pipelines and GitHub Actions; Terraform or Bicep Quality Vitest or Jest, Playwright, axe-core accessibility checks in CI, Storybook, contract tests, and LLM evaluation regression suites AI dev tooling Claude Code, GitHub Copilot, Cursor; repository context files (CLAUDE.md, AGENTS.md); custom skills, subagents, and MCP servers wired to the issue tracker, wiki, repositories, and cloud control planes What you will do Build the product end to end • Design and ship features across the full stack — React Server and Client Components, Server Actions and route handlers, the services and data model behind them, and the deployment that carries them to production. • Turn OpenAPI contracts into typed clients and validated boundaries, replacing prototype mock data with real integrations without degrading the interaction design the prototype established. • Extend and maintain a tokenized design system: accessible headless primitives, consistent interaction states, and components that behave correctly under keyboard, screen reader, reduced motion, and high contrast. • Own performance and reliability — streaming and Suspense boundaries, caching strategy, cold-start behavior, and error handling that degrades gracefully when a model call fails or times out. • Build and maintain integrations with enterprise systems of record: OAuth flows, webhooks, rate-limit-aware synchronization, and reconciliation when a remote system disagrees with local state. Engineer the AI layer • Design retrieval pipelines: chunking strategy, embeddings, hybrid search, reranking, and the freshness and permission model that keeps a retrieved document from leaking across tenants or roles. • Write and version prompts as engineered artifacts — schema-constrained outputs, tool and function definitions, few-shot context assembly — and treat context assembly as a design problem, not string concatenation. • Build tool-calling and agentic flows with real boundaries: idempotent tools, explicit confirmation gates on irreversible actions, timeouts, retries, and audit trails. • Stand up evaluation infrastructure — golden datasets, rubric and LLM-as-judge scoring, regression suites that run in CI — so a prompt or model change is a measured change rather than a hope. • Implement guardrails appropriate to sensitive and regulated data: PII detection and redaction, grounding and citation enforcement, refusal and escalation paths, and prompt-injection resistance on any surface that reads untrusted content. • Instrument cost and latency — token accounting, caching, model routing and fallback — and keep those tradeoffs visible rather than buried. Use AI tooling across the development lifecycle • Discovery and design. Use AI to interrogate an unfamiliar codebase, reconstruct undocumented behavior, and pressure-test an architecture before you commit to it. • Specification. Work spec-first: write the plan, the contract, and the acceptance criteria the agent will build against, and review that plan the way you would review a pull request. • Implementation. Drive agentic coding tools (Claude Code, Copilot, Cursor) with well-maintained repository context — CLAUDE.md or AGENTS.md files, custom skills, subagents, and MCP servers connected to the issue tracker, wiki, repositories, and cloud control plane. • Testing. Generate test harnesses rapidly and with quality — unit, contract, Playwright end-to-end, and axe accessibility assertions — and know the difference between coverage and confidence. • Review and refactor. Use AI for first-pass review, large-scale migration, and dependency upgrades, while remaining personally accountable for everything that merges. • Documentation. Produce runbooks, architecture decision records, and technical documentation as a byproduct of the work rather than a separate project. • Judgment. Know where these tools are weak — silent hallucination in unfamiliar APIs, confident wrong refactors, plausible tests that assert nothing — and verify accordingly. We do not ship unverified generated code, and we do not pretend the tools removed the need to understand the system. Required qualifications Full stack engineering • Five or more years building and shipping production web applications, with meaningful ownership of both client and server. • Deep proficiency in TypeScript and modern React, including Server Components, Suspense and streaming, and the data-fetching and state boundaries that come with the App Router. • Production experience with Next.js 13+ App Router, or equivalent depth in another modern full-stack framework plus a credible plan for coming up to speed quickly. • Strong CSS and design-system fluency — Tailwind or an equivalent utility or token system, and the ability to build components that hold up across themes, breakpoints, and user preference settings. • Backend competence in Node.js and at least one of Python, C#, or Go; REST and OpenAPI design, relational data modeling, and query performance. • Cloud delivery experience, preferably Azure: containerized deployment, managed identity and secret handling, structured logging and tracing, and CI/CD you configured yourself. • Authentication and authorization in an enterprise context: OIDC or SAML single sign-on, role-based access control, and multi-tenant data isolation. AI and LLM application engineering • Demonstrated production work with LLM APIs — Azure OpenAI, Anthropic, or OpenAI — beyond a demo or a chat wrapper. • Hands-on retrieval-augmented generation: you have chosen a chunking strategy, tuned retrieval quality, and dealt with the failure modes that follow. • Structured outputs and tool calling: JSON schema constraints, function definitions, and the validation and repair logic that keeps a malformed response from reaching a user. • A real evaluation practice — you can describe how you knew a prompt or model change improved something, with evidence rather than impression. • Streaming user interfaces: token streaming, partial rendering, cancellation, and accessible live-region updates that inform screen reader users without interrupting them. • Working understanding of AI safety and governance in an enterprise setting: data residency, PII handling, model and prompt versioning, human-in-the-loop design, and auditability. AI-assisted development practice • Daily, fluent use of at least one agentic coding tool, with a point of view on where it helps and where it does not. • Experience configuring the context these tools depend on — repository instruction files, custom commands or skills, and MCP server connections to real systems. • A track record of using AI to build meaningful test coverage quickly on a codebase that had none. • The discipline to review generated code as adversarially as you would review a stranger's pull request. Accessibility and quality • Practical WCAG 2.1 or 2.2 AA experience: semantic HTML, ARIA used correctly and sparingly, keyboard navigation, focus management, and testing with an actual screen reader. • Familiarity with automated accessibility testing in CI, and clarity about what automated checks cannot catch. • Testing habits that survive contact with a deadline: meaningful unit tests, end-to-end coverage of critical paths, and contract tests at integration boundaries. • Code review, version control, and release practices that hold up on a small team where you are often the last set of eyes. Preferred qualifications • Experience integrating with enterprise HR, ERP, or ITSM platforms — Workday, ServiceNow, Cornerstone, UKG, Salesforce, or NetSuite. • Embedded or extension surfaces: iframe and side-panel applications, host-application SDKs, or marketplace listings. • SCIM provisioning and enterprise customer onboarding at scale. • Experience testing with assistive technology users, or personal experience using assistive technology. • Prior work in a regulated or compliance-sensitive domain where an incorrect output has consequences for a real person. • Familiarity with the NIST AI Risk Management Framework, the EU AI Act, or comparable governance frameworks. • Data engineering adjacency — Azure Data Factory, Microsoft Fabric, Snowflake, or medallion-architecture pipelines feeding an application layer. • Contributions to open source, published technical writing, or conference speaking. What success looks like In the first 30 days — you have the application running locally, understand the architecture and the accessibility decisions behind it, have your AI tooling configured against our repositories, and have shipped a real feature to a staging environment. By 60 days — you own a vertical slice end to end: an API contract, the service behind it, the screens in front of it, and the tests and evaluations that cover it. You have replaced at least one mocked surface with a live integration. By 90 days — you are setting patterns rather than following them: retrieval and evaluation infrastructure others build on, an accessibility and testing standard enforced in CI, and the technical credibility to be in the room when the roadmap is decided. To apply Submit a résumé along with a short note covering two things: a production system you built where a language model was doing real work, and how you actually use AI tooling in your development process — what you delegate to it, what you refuse to, and how you verify the difference. Links to code we can read are worth more than a longer résumé. [Employer] is an equal opportunity employer. We evaluate all qualified applicants without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, age, veteran status, disability, or any other legally protected characteristic. Reasonable accommodations are available at every stage of the hiring process on request.