About This Developer

Krisztián Szalai

Full-stack engineer · AI-native · React Native

Currently open to new projects · Budapest, HU

I'm Krisztián Szalai, from Budapest, Hungary. I am a seasoned senior frontend / full-stack engineer with 7+ years of experience shipping web and mobile applications for startups, scale-ups, and enterprise teams. What sets me apart is strong UX and planning skills on both architecture and business logic — plus AI-native engineering: I orchestrate agentic workflows with Cursor and Claude Code, write specs before code, and review every agent output like a PR.

7+Years

7Projects

2Flagships

RNSpecialty

Flagship work

Projects — Finder

Favorites

AI-Assisted Development

AI-native engineering — not vibe coding

I design systems where agents are first-class participants: precise specs, curated context, deterministic verification, and human authority at merge. The bottleneck moved from typing to orchestration — I own that layer.

4Pillars mastered

SpecBefore code

HITLAt every gate

Agentic development lifecycle

Research
Spec
Delegate
Verify
Ship

AI-native principles

Spec-driven development

Every significant task starts with acceptance criteria, constraints, and edge cases — the shared interface between humans and agents (aligned with Microsoft's SDD and industry ADLC practice).

Context engineering

Rules, skills, AGENTS.md, and MCP servers shape the harness around the model. Two teams using the same LLM get different results — the environment is the edge.

Orchestrator–worker pattern

I stay on the critical path; parallel subagents explore, research, and draft. Single-responsibility agents with explicit handoffs — not one mega-prompt.

Evaluator–optimizer loops

Agent output is never rubber-stamped. I review like a PR: lint, typecheck, tests, and acceptance criteria must pass before merge — the executable definition of done.

Human-in-the-loop governance

Clear authority boundaries: agents draft and explore; I decide architecture, security-sensitive changes, and what ships. Escalation thresholds are explicit, not improvised.

Scoped decomposition

Large refactors split into verifiable chunks with rollback points. Deterministic orchestration on critical paths; agentic reasoning reserved for judgment-heavy subtasks.

Proof in production

KrisztiánOS — this portfolio7 projects · 3 locales
Rebuilt the entire site as a desktop-metaphor OS with agentic workflows: draggable windows, 3-locale i18n dictionaries, cross-linked project↔career graph, and an AI playbook window — scoped chunks with lint gates between each.
Morrison Street monitoring platformNext.js · CDK · PWA
Shipped multi-tenant Next.js features (RBAC, PWA, i18n, AWS CDK) using plan-first agent sessions — parallel exploration for unfamiliar APIs, atomic commits per feature slice.
Custom rules & skills library20+ domain skills
Authored reusable Cursor rules and skills for Prisma schemas, Cloudflare Workers, React Native profiling, and security review — encoding team conventions so agents don't repeat mistakes.
MCP integrations in daily workflow6+ MCP servers
Connected PostHog, Sentry, Prisma, Stripe, and browser DevTools via MCP — agents query live schema, docs, and CI status instead of hallucinating APIs.
Verification-first agent loopsLint · test · type gates
Agents run typecheck, lint, and Jest before I review. Test plans written first on risky changes — TDD reincarnated for the agentic era.
Git-aware multi-agent sessionsAtomic commits always
Branch-per-feature with Claude Code and Cursor Agent: plan mode maps blast radius, background agents explore while I hold architecture decisions, never freestyle on main.

Four pillars

New Message

Let's Talk

Bring a rough concept or a full spec — I'll reply with a plan. Currently open to new projects.

✉️ contact@krisztianszalai.com

GitHub ↗LinkedIn ↗