Order Management System

Date:

2026-03

Tech Stacks:

React
Node.js
MongoDB
Tailwind CSS
AWS S3
JWT

Overview:

Full-stack MERN admin platform for managing products and orders from Facebook and Instagram ads

Built a scalable MERN stack order management system designed for Facebook and Instagram ad-driven e-commerce. Implemented JWT authentication, RESTful APIs with Node.js and Express, and MongoDB schema design for products, orders, and users. Created an admin dashboard with order workflow management, and product CRUD functionality. Used AWS S3 for image uploads and Tailwind CSS for responsive UI. Structured the application for maintainability and scalability under high traffic conditions.

GitHub Repo

Authentication Service

Date:

2025-12

Tech Stacks:

Golang
Gin
PostgreSQL
JWT
Docker
OAuth2.0

Overview:

User Authentication and Authorization Server for microservices

Developed a Go microservice for user management using Gin framework, PostgreSQL, and JWT authentication. Provides REST endpoints: /api/register (user creation with bcrypt hashing), /api/login (JWT issuance), and authentication/authorization with Google OAuth 2.0 /api/me routes (GET/PUT/DELETE for self-profile) plus internal /api/users/:id. Uses middleware for JWT validation, env-based config, and Docker Compose for PostgreSQL

GitHub Repo

Instruction-Tuned Language Model

Date:

2025-08

Tech Stacks:

Deep Learning
HugingFace
PyTorch
AWS
OpenAI API

Overview:

Instruction-tuned large language model specifically for the Uzbek language by fine-tuning Mistral-7B using QLoRA

Created a high-quality synthetic dataset of ~23,000 diverse Uzbek instruction-response pairs (covering tasks such as summarization, description, question generation, calculation, translation, reasoning, and more) by leveraging GPT-4o through the OpenAI API at a total cost of only $88. Performed memory-efficient fine-tuning on a single AWS g5.xlarge instance (NVIDIA A10G GPU) with a compute cost of ~$15, demonstrating accessible and cost-effective adaptation of open-weight models to underrepresented languages. Evaluated the model across multiple language tasks to confirm improved instruction alignment, coherence, and task performance compared to the base Mistral-7B, contributing to the growing ecosystem of multilingual and low-resource LLMs.

Dataset is available on HuggingFace

Weather Forecast App

Date:

2026-01

Tech Stacks:

React
TypeScript
TanStack Query
Tailwind CSS
FSD Architecture
Vercel

Overview:

Korea Location-Based Weather Dashboard

Built a responsive South Korea weather app using React (functional components), TypeScript, Feature-Sliced Design (FSD) architecture, and TanStack Query for server-state management. On mount, it obtains the user geolocation via the browser API and fetches current temperature, daily min/max, and hourly forecasts from OpenWeatherMap or 공공데이터포털 API. Search supports any Korean administrative unit (시/도/구/동) by indexing the provided korea_districts.json for prefix/autocomplete matching, mapping selected entries to lat/lon or grid coordinates, with fallback UI for missing weather data. Favorites (max 6) are persisted in localStorage, support editable aliases, and render as cards showing current conditions + min/max; clicking opens a detail page with full forecast data. Tailwind CSS handles mobile-first responsive layout. All async operations are managed declaratively via TanStack Query with proper caching, error handling, and optimistic patterns where applicable.

Live Demo