Available for Hire · E-7 Visa · South Korea

Makhmud
Kudratov

// aka Mike  ·  github.com/MakhmudSD

AI / Agentic AI Engineer with 4+ years of software engineering experience across EdTech, e-commerce, and SaaS — LangGraph agent orchestration, RAG pipelines, and document intelligence systems, shipped end-to-end as a full-stack developer owning both backend APIs and frontend delivery. Led a 5-person engineering team in South Korea, driving client delivery and AI-driven feature development.

Suwon, Gyeonggi-do Remote · US / Europe / Asia +82 10-9380-7522 makhwork15@gmail.com
Makhmud Kudratov

About

About Me

I'm an AI / Agentic AI Engineer based in Suwon, South Korea, with full-stack ownership from backend APIs to frontend delivery. Over the past 4+ years I've built production systems across EdTech, e-commerce, and SaaS — full-stack engineering throughout, with the most recent stretch focused on RAG pipelines, document intelligence platforms, and AI feature integration, including two AI products I designed and built solo end-to-end.

I'm currently an AX Engineer at Langcode, working across the full stack on a Fortune Global 500 semiconductor manufacturer's internal AI Transformation platform — contributing to a Magentic-based agent data pipeline, adding OpenTelemetry-based distributed tracing into the backend's Serilog pipeline on a C#/.NET stack, working with a separate FastAPI/Python OpenSearch service, and refactoring a frontend where logic was scattered across inline hooks into grouped, reusable custom hooks with centralized error handling.

Before that, at Ainex Lifetime Education, I was promoted from Developer to Team Lead, managing 5 engineers across 18 client maintenance projects. I handled LiveKit-based live classroom infrastructure, PayApp payment integrations, and Apache server operations — while keeping client satisfaction high and systems stable.

Earlier in Korea, I worked as a Junior Frontend Engineer at Levelupsoft (Gwangju), maintaining a large food delivery platform, then as a full-stack developer at BTREE (Jeonju), building internal tools and client platforms with Next.js, NestJS, and GraphQL.

I hold an E-7 Skilled Worker Visa, am fluent in both Korean and English, and am open to remote or on-site engineering roles worldwide.

AX Platform — Agent Orchestration
Self-initiated project inspired by real enterprise approval workflows — LangGraph-orchestrated multi-agent request routing, live at agent-platform-sooty.vercel.app.
Lexara — AI Document Intelligence
Solo-built a full-stack RAG SaaS platform in under a week — 5-language UI, subscription tiers, live at lexara-top.vercel.app.
LLM & RAG Engineering
Built RAG pipelines, document intelligence systems, and AI feature integrations across client projects using OpenAI, pgvector, and LangChain.
Team Lead & Client Delivery
Led 5 engineers across 18 concurrent client projects at Ainex Lifetime Education — coordinating sprints, client communication, and stable production deployments.
Multinational Experience
Native Uzbek speaker, based in South Korea for 4+ years across Yeosu, Gwangju, Jeonju, Yongin, and Suwon. Fluent in Korean & English.
System Design & Architecture
Designed the hybrid BM25+pgvector retrieval and multi-agent orchestration flow for AX Platform, added distributed tracing across Langcode's backend, and built the real-time sync architecture (Supabase) behind Secret House's booking platform.

Technical Skills

Tech Stack

Frontend
React.jsNext.jsVue.jsAngularTypeScriptJavaScript ES6+ReduxSocket.ioThree.jsTailwind CSSShadCNFramer MotionSCSS
Backend
Node.jsNestJSExpress.jsPHPPythonGraphQLREST APIPrisma ORMJWTFastAPIJAVAC#.NETSignalRSerilogWebSocket
Databases
PostgreSQLMySQLMongoDBRedisSupabasepgvectorOpenSearch
AI & LLM
LangGraphLLM Orchestration (LangChain)RAG (Retrieval-Augmented Generation)Embeddings & Semantic SearchFunction Calling / Tool UsePrompt EngineeringAI Agent WorkflowsMagentic (Multi-Agent Orchestration)Context & Memory Management
Infrastructure & DevOps
Docker / Docker ComposeAWS (EC2 · S3 · CloudFront · Lambda · RDS)GitHub Actions CI/CDGitLab CI/CDNginx · ApachePM2CloudflareLinux UbuntuLiveKit
Security & Performance
OAuth 2.0 / OIDCOpenTelemetrySentry (Error Tracking)
Testing & API Tools
JestVitestReact Testing LibraryPlaywrightPostmanSwagger / OpenAPI
AI-Assisted Development
Claude CodeCursorOpenAI CodexFigma

Experience

Work Experience

AX EngineerJun 2026 – Present

Langcode · Enterprise AI Transformation (AX)

  • Frontend refactor: Inherited a component tree where business logic was scattered across dozens of inline hooks written directly inside components, making state changes hard to follow. Refactored by grouping related hooks into reusable custom hooks and implementing centralized error handling — removed duplicated error-handling code across views and made state logic traceable end-to-end.
  • Backend observability: Production issues were hard to diagnose from flat log lines alone. Integrated OpenTelemetry into the existing Serilog logging pipeline, adding distributed tracing on top of structured logs for real request-level visibility.
  • The platform's AI processing layer was limited to single-call LLM usage with no way to chain multi-step tasks. Contributed to a Magentic-based multi-agent data pipeline, extending the platform toward repeatable, multi-agent workflows for a Fortune Global 500 semiconductor manufacturer's internal AI processes.
  • Search: The platform needed full-text search beyond what simple database queries could handle. Implemented OpenSearch as an independent FastAPI/Python microservice to provide scalable full-text search.
React.jsTypeScriptC#.NETSignalRMagenticOpenTelemetrySerilogPostgreSQLREST APIOpenSearchMongoDBGitLabJira
Full-Stack Developer → Team LeadOct 2025 – Jun 2026

Ainex Lifetime Education · Yongin, Gyeonggi-do · On-site

  • Platform ownership & AI integration: The government assignment platform (4,000+ monthly users, 13 client accounts) had no single owner — feature requests and bug fixes were falling behind, and Apache server config across those environments wasn't consistently maintained. Took over as primary owner: built and maintained 70%+ of the platform, shipped OpenAI API-driven features across multiple client accounts, and kept all 13 stable.
  • Performance under load: Page loads on the busiest course and content pages were slow. The cause: unoptimized MySQL queries on high-traffic tables, plus media files served directly from the origin server with no CDN in front of them. Rewrote the slow queries and added a CDN layer for media delivery — load times came back down on the heaviest-traffic pages.
  • Live classroom + payments: The platform had no way to run live video classes for corporate training, and no way to charge for them. Designed the backend APIs for a LiveKit-based live classroom system and integrated PayApp server-side end to end (webhook handling, payment verification, transaction management), giving the platform its first working path from a scheduled training session to a verified, paid enrollment.
  • Team Lead promotion: The client base grew past what one developer could support alone, so I was promoted from Developer to Team Lead. Managed a 5-person engineering team across 18 concurrent client maintenance projects (the 13-account assignment platform included) — sprint planning, task distribution, and client communication — while keeping client satisfaction and delivery quality high.
Next.jsReact.jsNode.jsPHPPythonOpenAI APIApacheLiveKitPayAppMySQLDocker
Full-Stack DeveloperJul 2023 – May 2025

BTREE · Jeonju, South Korea  ·  Software Development & IT Consulting

  • Full-stack feature delivery: Joined as a full-stack developer building internal tools and client-facing platforms with no single point of ownership across the stack. The team needed consistent Next.js/NestJS/MongoDB delivery without waiting on a dedicated architect for every feature. Built and maintained production features end-to-end, designing GraphQL schemas and REST APIs and optimizing large-scale backend queries with Prisma ORM as data volume grew.
  • Auth & deployment pipeline: Services had inconsistent auth and deploys were manual, which stopped scaling as the number of services grew. Needed stateless auth that worked across distributed services and a CI pipeline that removed the manual deploy step. Implemented JWT-based stateless auth, containerized services with Docker, and set up GitHub Actions CI/CD pipelines for automated builds and deploys.
Next.jsNestJSMongoDBGraphQLPrisma ORMDockerGitHub ActionsTypeScript
Junior Frontend EngineerApr 2022 – May 2023

Levelupsoft Inc · Gwangju, South Korea  ·  Digital Solutions

  • Cross-browser & mobile fixes: The frontend of a large food delivery platform rendered inconsistently across browsers and lagged on mobile, directly affecting real paying customers. Needed reliable cross-browser behavior and better mobile performance without a full rewrite. Maintained and improved the frontend, fixing cross-browser compatibility issues and optimizing for mobile.
  • Design-to-code: Design and engineering worked from disconnected files, so Figma mockups tended to drift from what actually shipped. Needed a direct path from design to production UI to keep the two in sync. Translated Figma designs into production-ready, responsive React components.
React.jsReduxJavaScriptSASSFigma
Frontend/Backend Developer — InternshipJan 2022 – Mar 2022

와이랩스 · Yeosu

  • University internship: Joined 와이랩스 (Yeosu) with no professional dev experience yet and needed hands-on practice turning requirements into working code. Learned frontend/backend fundamentals, implemented CRUD operations, and translated Figma designs into responsive UI.
JavaScriptHTMLCSSREST APIFigma

Projects

Key Projects

AI Engineering · Agent Orchestration

AX Platform — Agent Orchestration System

View Live

A self-initiated agent orchestration platform, inspired by real enterprise approval workflows, where employees describe requests in plain language and a LangGraph orchestrator routes them through policy retrieval, LLM-driven drafting, and a three-tier role-based approval chain (Requester → Approver → Admin), with every state transition persisted to Postgres. Deployed live on Vercel/Render/Supabase with real per-call token/cost tracking. Root-caused and fixed a cross-site authentication cookie bug in production. Backend runs on Render's free tier; cold start after idle takes ~79s, confirmed via direct HTTP timing.

LangGraphFastAPISQLAlchemyGemini APIJWTbcryptPostgreSQLpgvectorWebSockettenacityOpenTelemetryLangSmithNext.jsTypeScriptTailwind CSSVercelRenderSupabase
Login
Login
Requester home — meet the agents
Requester home — meet the agents
Requester side — a real back-and-forth, not a form
Requester conversation — natural back-and-forth, not a form
Policy citations and routing rationale, generated live
Policy check — cited automatically, routed with a stated rationale
Approver side — full reconstructed transcript of the requester's conversation
Approver view — the full conversation that produced this request
Admin console — full audit trail and decision record
Admin console — full audit trail per request
Admin console — live token/cost tracking
Admin console — live token/cost tracking
Admin console — user management
Admin console — user management
Notification feed
Notification feed — every status change surfaced
Notification preferences, role-aware
Notification preferences — scoped to what your role can actually receive
Help — how the agents work, explained step by step
Standalone user guide — how each agent works, step by step
Why these technologies
LangGraph — Real manager/specialist node architecture with genuine loopback routing and a suspend/resume state machine, not a hardcoded pipeline.
pgvector + BM25 (RRF) — Keyword and semantic retrieval fused via Reciprocal Rank Fusion for policy lookup during drafting, without standing up a separate vector database.
WebSocket — Request status and approval events push to requester/approver/admin clients over a persistent connection instead of polling on an interval.
JWT + bcrypt — Stateless, role-aware session auth across requester, approver, and admin roles, with passwords hashed before storage.
OpenTelemetry + tenacity — Distributed tracing across the FastAPI backend paired with automatic retry/backoff on LLM and DB calls, so transient failures don't surface as user-facing errors.
Full-Stack · Hospitality Booking

Secret House — Private Villa Booking Platform

Freelance-built booking platform for a private luxury villa business in Yeosu — replaces manual phone/DM bookings with a self-serve reservation flow. Real-time availability calendar backed by Supabase, a Toss Payments deposit flow to cut no-shows, trilingual UI (EN/KO/UZ) via next-intl for both local and international guests, and a full admin console (booking control, occupancy/RevPAR analytics) with instant Telegram alerts plus email notifications the moment a new reservation comes in.

Next.jsSupabaseToss Paymentsnext-intlTailwind CSSTelegram Bot APIVercel
Homepage — hero, live villa photography
Homepage — hero, live villa photography
Gallery — inside the villa (private pool)
Gallery — inside the villa
Amenities — pool, sauna, 3 bedrooms, dual kitchens, triple view
Amenities — everything the listing promises, itemized
Transparent tiered pricing by day of week
Transparent pricing — weekday/Friday/weekend tiers
Live availability calendar
Live availability calendar
Why these technologies
Supabase — Managed Postgres + realtime subscriptions kept the availability calendar in sync across the booking form and admin console without a custom polling layer.
Toss Payments — KRW-native payment rails matched the primarily domestic guest base; a deposit-based flow reduced no-show risk on a single-unit property.
next-intl — Trilingual routing (EN/KO/UZ) served both local Korean guests and international travelers without duplicating page logic per locale.
Full-Stack · E-Commerce

Veloura Jewelry

Full-stack jewelry marketplace with real-time WebSocket notifications, multilingual support (Korean, English, Uzbek), GraphQL-powered dynamic URL filtering, built-in payment cart, community features, and a comprehensive admin panel. Deployed on Vercel (frontend) and Railway (backend).

Next.jsNestJSGraphQLMongoDBWebSocketDockerTypeScript
Veloura JewelryVeloura Jewelry Mobile
Why these technologies
NestJS — Built-in DI container structured the codebase to reduce merge conflicts as the team scaled features.
GraphQL (Apollo) — A single endpoint with flexible queries replaced 15+ REST endpoints; the frontend fetches only the exact fields needed per view.
MongoDB — Jewelry product attributes vary per type (rings vs necklaces); flexible schema modelling was more natural than rigid SQL tables.
WebSocket — Real-time likes/follow notifications required persistent connections; polling would have created excessive DB reads.
Docker Compose — Spun up the API and batch worker together locally with one command, keeping the dev environment consistent before deploying to Railway.
Full-Stack · E-Commerce

Veloura Perfume

E-commerce platform for a local perfume store with store registration, product management, order tracking, and Redux Toolkit cart. Includes a separate admin panel built with EJS and Node.js. Deployed on Linux VPS with PM2 and Nginx.

React.jsTypeScriptExpress.jsMongoDBRedux ToolkitPM2Nginx
Veloura Perfume OrdersVeloura Perfume Shop
Why these technologies
React + TypeScript — Strict typing across product and order interfaces caught checkout bugs at compile time before they reached users.
Express.js — Simpler REST scope didn't justify NestJS overhead; Express kept the codebase lean and deploy-ready in days.
Redux Toolkit — Cart state needed to survive route changes and component unmounts; RTK's immer-based reducers kept mutations predictable.
PM2 — Cluster mode spread the app across all VPS CPU cores instead of running on one; auto-restart kept the store online after any uncaught exception.
Nginx — Reverse-proxied the PM2 port to the domain with HTTPS termination—standard production pattern for Node.js on Linux VPS.

Education & Languages

Education & Languages

Bachelor's Degree
Chonnam National University · Gwangju
2019 — 2023
42.uz — Programming Fundamentals
Intensive project-based course, algorithms & software fundamentals · Uzbekistan (online)
Korean
Fluent
English
Fluent
Russian
Conversational
Uzbek
Native

Contact

Let's Work Together

Open to remote full-stack or team lead roles with US, European, and Asian teams. Feel free to reach out anytime.

E-7 Skilled Worker Visa · South Korea · Available for Global Remote Work