Open to relocation · Netherlands / Ireland / Canada · sponsorship-ready

I build the layer where product engineering meets LLMs.

14 years shipping systems that couldn't go blind — The New York Times, an Australian investment bank's deal platform, a crypto exchange order book, an education platform used by 1.8M learners. Now: AI-native developer platforms, RAG and agents in production on AWS Bedrock.

Hiring me See AI projects Get in touch

14+years

1.8Mlearners

$200M+daily flow

~30engineers ship on the SDK

~10Kdocker pulls

portfolio.agent

agent.tracesession · 00/09live

›userHas Arif worked on AI agents in production?

model

gpt-4o

tools

4 active

latency

~480ms

Next.js 16◆TypeScript◆LangGraph◆Tavily Search◆Upstash Redis◆React 19◆Framer Motion◆Tailwind v4◆Shadcn UI◆Streaming UX◆ReAct◆Tool use◆Evals◆MCP◆stdio MCP server◆Docker◆npm◆Nuxt 3◆LangGraph.js◆Fireworks.ai◆Pinia◆Multi-agent◆AWS Lambda◆Bedrock Guardrails◆Claude Haiku◆NestJS◆DynamoDB◆AWS CDK◆CloudFormation◆CloudFront◆ECR◆Prompt firewall◆Ports & adapters◆AI code review◆Repo-aware review◆GitHub App◆Claude on Bedrock◆Octokit◆Next.js 16◆TypeScript◆LangGraph◆Tavily Search◆Upstash Redis◆React 19◆Framer Motion◆Tailwind v4◆Shadcn UI◆Streaming UX◆ReAct◆Tool use◆Evals◆MCP◆stdio MCP server◆Docker◆npm◆Nuxt 3◆LangGraph.js◆Fireworks.ai◆Pinia◆Multi-agent◆AWS Lambda◆Bedrock Guardrails◆Claude Haiku◆NestJS◆DynamoDB◆AWS CDK◆CloudFormation◆CloudFront◆ECR◆Prompt firewall◆Ports & adapters◆AI code review◆Repo-aware review◆GitHub App◆Claude on Bedrock◆Octokit◆

01 · Current AI work

Production AI engineering

Shipping today. Open source where I can.

● IN PRODUCTIONcurrent

AI-First Development SDK

AI-First SDK · an energy company (client under NDA)

Migrated a monolithic industrial control application into a modular, AI-friendly SDK — documentation, CLI tooling for LLM agents, and an MCP server exposing component metadata — so ~30 engineers and their models compose UI against real contracts instead of guessing. The LLM operates the station and generates UI on the fly from a UI kit with strict prop contracts. In daily use.

Skills demonstrated

›AI-friendly SDK design
›MCP server for component metadata
›CLI tooling for LLM agents
›Machine-readable component contracts
›Monolith → modular migration
›Leading the client frontend team

Stack

TypeScriptMCP serverCLI for LLM agentsUI kit + prop contractsComponent metadataAgent skills

Signals

◆In daily use by ~30 engineers
◆Monolith → modular SDK
◆I lead a frontend team of 4–6 and still write code daily

Read the case study

● LIVE2026

Nexus Sentinel

Prompt Firewall · AWS + Bedrock

A self-hosted prompt firewall for any LLM. One endpoint — POST /v1/verify — takes a prompt plus a policy and returns a structured allow / redact / block verdict with matched categories, confidence scores, a redacted preview, and a replayable audit log. Built on Amazon Bedrock Guardrails + Claude Haiku with a ports-and-adapters core, so the full test suite and a live demo run offline against deterministic fakes — no AWS account required.

Skills demonstrated

›LLM safety & guardrails
›Prompt-injection defense
›Ports & adapters architecture
›AWS Lambda + API Gateway
›Bedrock Guardrails + Claude Haiku
›Layered rate limiting & cost caps

Stack

TypeScriptNestJSNext.js 16AWS CDKAmazon BedrockClaude HaikuDynamoDBLambdaCloudFrontzod

Signals

◆Lighthouse 100 × 4
◆Runs fully offline (fakes)
◆Live on AWS · CI-gated

Open live demo How it works API reference View source

● IN PRODUCTIONcurrent

AI Code Review, Zero to Production

An AI code-review GitHub App · Claude on Bedrock

A GitHub App that reviews every pull request in the context of the whole repo, not just the diff. A native Claude-on-Bedrock tool-use loop walks the changed files, their neighbours and their tests through the GitHub API, checks them against a language-agnostic rubric, and posts small, line-anchored comments. GitHub App → AWS Lambda → Claude on Amazon Bedrock, infrastructure as code with CDK and least-privilege IAM, and an offline eval harness that gates every prompt and model change on recall, precision and stability.

Skills demonstrated

›Agentic tool-use loop
›Repo-aware code review
›Claude on AWS Bedrock
›GitHub App + webhooks
›Eval harness gating prompt & model changes
›IaC with CDK · least-privilege IAM

Stack

TypeScriptAWS LambdaAmazon BedrockClaudeGitHub AppOctokitAWS CDKZodesbuild

Signals

◆Zero to production
◆Single Lambda · no DB
◆Eval-gated prompt & model changes

Read the case study

● LIVE2026

Nexus Recall

RAG Document Explorer

Browser-first RAG document explorer with PDF and Markdown ingestion, local embeddings, IndexedDB vector search, highlighted citations, and streaming AI answers powered by Fireworks.ai, OpenAI, and Claude through the Vercel AI SDK.

Skills demonstrated

›RAG architecture
›Local-first embeddings
›Vector search
›Citation extraction
›Streaming AI chat
›SvelteKit PWA

Stack

SvelteKitSvelte 5TypeScriptVercel AI SDKFireworks.aiOpenAIClaude / AnthropicTransformers.jsLangChain.jsIndexedDB

Signals

◆Browser-first RAG
◆Local embeddings
◆Open source

Open live demo How it works View source

● LIVE2026

Nexus Trace

Streaming AI Agent

Streaming AI agent with live tool-call trace visualization. Ask a question, watch the model search the web, call tools, and reason through its answer in real time. Built with a ReAct loop, SSE token streaming, and a real-time trace UI.

Skills demonstrated

›Agent orchestration
›Tool-call streaming
›ReAct pattern
›SSE / token streaming
›Trace observability UI
›Next.js Server Components

Stack

Next.js 16TypeScriptLangGraphTavily SearchUpstash RedisTailwind CSS v4

Signals

◆Real-time trace
◆Multi-tool reasoning
◆Open source

Open live demo How it works View source

02 · Field record

14 years of production systems

Each one reframed for what it teaches about AI engineering at scale.

01● LIVE

A crypto exchange (name under NDA) · 2022–2024

Crypto Exchange Order Book

Real-time markets under a strict render budget

Order-book UI for a crypto exchange (name under NDA), from scratch to production — real-time market data at high update frequency, virtualization, a strict render budget, and stream handling kept outside the React tree. Then broad feature work and production debugging across the platform.

ReactTypeScriptReal-time streamsVirtualization

Zero to production

High-frequency market data

02● LIVE

An energy company

Power-Plant Control & Reporting

Realtime telemetry → LLM-ready operations

Operator-facing control and reporting for a power plant — hundreds of live sensors, real-time telemetry, and dashboards that are not allowed to go blind or lag.

ReactTypeScriptWebSocketsRTK QueryThree.js

Hundreds of live sensors

Real-time telemetry

03● LIVE

Avallain · 2019–2022

Learning Platform for 1.8M Learners

AI-ready learning infrastructure

Joined as a frontend developer, grew into the technical lead, and rebuilt a legacy platform used by 1.8M learners for one of the world's largest education publishers (client under NDA) — refactor, automated testing, a modernised stack, and the customer in every Scrum call. The company had written the project off as non-profitable; it became its main income stream.

ReactTypeScriptRuby on RailsPostgreSQL

1.8M learners

Became the company's main income stream

Read the case study →

04● LIVE

Luxoft · 2016–2018

Investment Bank Deal Platform

Compliance-grade UI for high-stakes data

Deal-processing platform for one of Australia's largest investment banks, $200M+ of daily deal flow. Led 6 teams, 30 engineers and QA across 4 time zones; grew a team 5× in six months while staying top contributor by commits and merged PRs. Owned hiring, onboarding, Scrum, environments and CI — the same rigor I now apply to LLM evals and guardrails.

React.jsReduxAngularJestREST APIs

$200M+ daily flow

6 teams · 30 engineers

Grew a team 5× in six months

Read the case study →

05● LIVE

EPAM Systems / The New York Times · 2015–2016

NYT Marketing Tools & A/B Testing

Experiment instrumentation at scale

Internal tools for the New York Times marketing team, and the A/B testing experiment framework — designing, instrumenting and analysing experiments to grow reader engagement. The original data-driven UX work that now translates directly to LLM evals.

React.jsAngular.jsNode.jsMongoDB

A/B testing framework

Experiment instrumentation

06● ARCHIVED

WINGS Stiftung · 2018

DAO Creation & Funding Platform

On-chain state, by hand

One of the early DAO creation and funding platforms — keeping UI state in sync with on-chain state by hand, years before wagmi/viem and server-state libraries existed.

AngularWeb3SolidityEthereum

Smart contracts

DApps

03 · Relocation

How hard is it to hire me?

Short answer: a standard permit, employer-initiated, no degree requirement at this band.

Where

Netherlands (primary), Ireland, Canada

Permit

Dutch highly-skilled migrant or Irish Critical Skills — salary/experience route, no degree requirement

Spouse

Work-authorized on the dependent permit from day one

Working style

Hybrid or on-site is fine — no remote-only requirement

Full relocation details

04 · Capability map

What I bring to a team

AI Engineering

Focus

LLM-powered appsAgentic workflowsAgentic code reviewRepo-aware code reviewGitHub App automationRAG / retrievalLocal-first embeddingsVector searchSemantic chunkingStructured citationsTool use & function callingModel Context Protocol (MCP)MCP serversStreaming UXPrompt engineeringEvals & guardrailsLLM eval harnessesLLM safetyPrompt-injection defensePrompt firewallsEmbeddings pipelinesMulti-agent systemsVercel AI SDKLangGraph.jsLangChain.jsTransformers.jsFireworks.aiOpenAIClaude / Anthropic

Frontend Development

Expert

React.jsNext.jsSvelteKitSvelte 5Vue 3Nuxt 3TypeScriptJavaScriptReduxTanStack QueryZustandPiniaReact NativeHTML5CSS3Three.js

UI/UX Development

Expert

Tailwind CSSSassStyled ComponentsFramer MotionResponsive DesignDesign SystemsUser Interface Development

Backend Integration

Advanced

Node.jsExpressNestJSREST APIsGraphQL.NET C#

Cloud & AWS

Advanced

AWS LambdaAPI GatewayAmazon ECRDynamoDBAmazon BedrockBedrock GuardrailsCloudFrontS3SSM Parameter StoreIAM (least-privilege)CloudWatchAWS BudgetsAWS CDK (IaC)CloudFormationServerless architecturePorts & adaptersCost & rate limiting

Web3 Technologies

Advanced

SolidityBlockchain DevelopmentDAppsSmart Contracts

Database Systems

Intermediate

PostgreSQLMongoDBIndexedDBVector StoresMS SQLDatabase Design

Testing & Quality

Advanced

PlaywrightCypressJestTesting LibraryVitestTest-Driven Development

DevOps & Deployment

Advanced

CI/CDGitHub ActionsDockerVercelCloud Infrastructure

Performance Optimization

Expert

Core Web VitalsLighthouseBundle Size OptimizationWeb Performance

Leadership

Expert

Team LeadershipAgile/ScrumProject ManagementTechnical MentorshipCode Reviews

Languages

Native

English (Full Professional)Russian (Full Professional)

05 · Credentials

Verified

Recently formalized — alongside 14 years of production work.

DataCamp

verified · 2026

AI Engineer for Developers

credential · AIEDA0019287935520

LLM AppsRAGTool UseAgentsEmbeddings

view ↗

06 · Studio

Demos & creative experiments

Weekend builds and experiments. The production work is in the Field Record above.

● LIVE

Nexus Sentinel

Self-Hosted Prompt Firewall for Any LLM

POST a prompt plus a policy to /v1/verify and get a structured allow / redact / block verdict — matched categories, confidence scores, a redacted preview, and a replayable audit log. Bedrock Guardrails + Claude Haiku behind a ports-and-adapters core that also runs fully offline against deterministic fakes.

TypeScriptNestJSNext.js 16AWS CDK

Live on AWS

allow / redact / block

Runs offline (fakes)

Live demo How it works API Source

● LIVE

Nexus Forge

AI Mind Map with Live Agent Reasoning

Infinite-canvas mind-map with four distinct AI personalities (AXIOM-9, VERN, ORACLE-3, PATCH) that analyze your thinking, stream dry reasoning live, and restructure your map with a click. Custom SVG canvas, SSE streaming, LangGraph workflow.

Nuxt 3Vue 3TypeScriptLangGraph.js

4 AI Personalities

Live SSE Reasoning

Open Source

Live demo How it works Source

● LIVE

Nexus Recall

Browser-First RAG Document Explorer

Upload PDFs and Markdown, embed documents locally in the browser, search chunks through IndexedDB vector retrieval, and ask streaming RAG questions with structured citations and document highlighting.

SvelteKitSvelte 5TypeScriptVercel AI SDK

PDF + Markdown RAG

Local embeddings

Streaming citations

Live demo How it works Source

● LIVE

Nexus Trace

Streaming AI Agent

Streaming AI agent with live tool-call trace visualization. Ask a question, watch the model search the web, call tools, and reason through its answer in real time.

Next.js 16TypeScriptLangGraphTavily Search

Real-time trace

Multi-tool reasoning

Open source

Live demo How it works Source

● LIVE

stdio Context7 MCP

Published MCP Server · npm + Docker

A published Model Context Protocol server that streams up-to-date library docs and code examples into any MCP client — Claude, Cursor, Windsurf. stdio transport, native Docker MCP Toolkit support, multi-arch images, and non-root hardening. Distributed on npm and Docker Hub under @dolasoft.

TypeScriptNode.js ≥20.8MCP SDKDocker

~10K Docker pulls

npm + Docker Hub

Open source · MIT

npm Docker Source

07 · Writing

Field notes

AI Engineering

Evaluating RAG Systems: Beyond Vibes

How to stop guessing and start measuring what actually matters

2026-05-204 min readRead →

AI Engineering

Streaming AI Responses: SSE vs ReadableStream vs Vercel AI SDK

How I stopped waiting for walls of text and made my AI feel alive

2026-05-205 min readRead →

AI Engineering

The Kingdom of Shattered Memories

Why Your RAG Forgets Where It Left Its Brain — and the Dark Spell That Fixes It

2026-05-216 min readRead →

AI Engineering

Building Your First Production-Ready MCP Server

A weekend project that actually shipped — to npm, Docker Hub, and real AI assistants

2025-10-026 min readRead →

08 · Open a line

Let's build something with AI.

Open to staff and lead roles in the Netherlands, Ireland and Canada — anywhere LLMs, agents, and great UX intersect. Sponsorship-ready, hybrid or on-site, comfortable across timezones.

Send email LinkedIn Download CV

arifdewi.uae@gmail.com

Phone / WhatsApp

+971 52 750 0822

linkedin.com/in/arifdewi

GitHub

github.com/arifdewiuae

t.me/arifdewi

Location

Dubai, UAE · relocating to the Netherlands / Ireland / Canada