A collection of personal and academic projects showcasing hands-on experience in machine learning, automation, cloud systems, and more. Explore my work in AI, ML, and software development
React • TypeScript • Supabase • TailwindCSS
BookmarkHub is a full-featured personal bookmark management application built for speed, privacy, and cross-device sync. The frontend is implemented in React + TypeScript and styled with TailwindCSS for a responsive, accessible UI. Users can create categories and tags, perform fast client-side search, and annotate bookmarks with descriptions and custom metadata. The app integrates with Supabase to provide optional cloud persistence; when no cloud connection is configured it gracefully falls back to using LocalStorage so users can continue working offline. The data model and import/export utilities make migration and backups straightforward. The codebase emphasizes modular components, unit-testing-friendly utilities, and clear separation between UI and storage adapters.
Architecture & Notes: Client-side React app with a thin sync adapter to Supabase. Local-first design, optimistic UI updates, and support for anonymous usage via ephemeral browser IDs. Good starting point for adding user auth, folder-like bookmark organization, or browser extension support.
Live Demo → View GitHub Repo →Azure • Python • Docker • OpenAI
Deep Chatbot demonstrates an end-to-end conversational AI system intended for production deployment. The architecture uses Azure Bot Service for channel-agnostic messaging, containerized model inference services for running large language models (or proxies to external OpenAI-style endpoints), and Azure AI Search as a vector index for persistent memory and semantic retrieval. The system integrates user profiles and conversation state to allow personalized responses, and it uses CI/CD pipelines to build, test, and deploy container images automatically. The codebase includes abstractions for fallback strategies (vector search first, text-based fallback), memory retention policies, and tools for importing documents into the search index.
Key details: modular microservices for inference and memory, secure secrets handling for API keys, and example pipeline configs for GitHub Actions / Azure DevOps. A good reference if you want to add retrieval-augmented generation and long-term memory to a bot.
View GitHub Repo →Next.js • Python • MCP • SerpAPI
ReminderBot is an AI-first event discovery and scheduling automation system. Its backend orchestrates a set of adapters that search the web (via SerpAPI and custom scrapers), then uses MCP-compatible AI extraction tools to convert messy HTML pages into structured event objects. Extracted events are normalized (dates, times, timezones, locations) and enriched with metadata like source title and confidence scores. The system can push events to arbitrary webhook endpoints such as Pipedream workflows to create calendar entries automatically, and it includes a Next.js frontend for configuring queries, viewing results, and managing webhook targets.
Architecture & features: modular fetcher/adapters, MCP tool registry for deterministic extraction, timezone-aware normalization logic, JSON output schema for easy calendar integration, and example Pipedream workflows. Good for automating calendar population and building tool-assisted extraction pipelines.
View GitHub Repo →Flask • Python • Pandas • NumPy
The Drone Decision Support System helps operators and planners choose the right unmanned aerial vehicles for specific port-related missions. It combines environmental data (wind, visibility, space constraints), mission requirements (range, payload, endurance), and a drone capability database to compute suitability scores. The backend is implemented in Flask and uses Pandas for dataset management and scoring logic. The frontend provides interactive forms to configure port parameters, select mission types, and inspect ranked recommendations along with rationale for each ranking.
Why it helps: automates multi-criteria selection, documents trade-offs (cost vs capability), and is easily extensible with new drone types and environmental models. Useful for research prototyping or operational decision support in maritime/port domains.
View GitHub Repo →Python • FastAPI • Kubernetes • MongoDB
GearboxGuard is an applied predictive maintenance platform that simulates and analyzes gearbox vibration signals to diagnose faults early. The project includes signal simulation tools, feature extraction services (time- and frequency-domain features), and an ML classification service trained on labeled vibration datasets. It's organized as microservices (FastAPI endpoints) to separate responsibilities: signal generation, feature extraction, classification, and result logging. Docker Compose and Kubernetes manifests make local testing and cloud deployment straightforward, and MongoDB is used for storing raw signals, computed features, and prediction history.
Notable components: Jupyter notebooks for model development, reusable feature extraction utilities, production-friendly model serving patterns, and Kubernetes HPAs for autoscaling during load testing. The system is a good foundation for predictive maintenance experiments and demonstrations.
View GitHub Repo →Python • Ollama • Hugging Face
This repository captures progress through an AI Agent course using Hugging Face ecosystems and local runtime tooling (Ollama). The project explores agent design patterns—tool usage, prompt engineering, turn-level context management—and experiments with integrating the qwen2:7b model for lightweight, local inference. Notes, reproducible environment scripts, and small agent prototypes that call external tools are included to demonstrate practical agent workflows.
Learning focus: agent orchestration, safe tool usage patterns, model selection trade-offs, and deploying small-scale agents to local or cloud runtimes.
View GitHub Repo →Next.js • Flask • Supabase • Tailwind CSS • QR Generation
A comprehensive event management platform that generates and manages QR codes for guest attendance tracking. Built with Next.js frontend and Flask backend, featuring real-time QR code generation, guest management, and event analytics. The system allows creating events, bulk uploading guests via CSV, automatic QR code generation containing guest contact information, and real-time attendance tracking through QR scanning. Includes a responsive neumorphic UI design, live dashboard with attendance statistics, secure Supabase database storage, and interactive data visualization.
Key Features: Event management, guest management with bulk upload, QR code generation, attendance tracking, dashboard analytics, responsive design, real-time updates, secure storage, API integration, and data visualization.
View GitHub Repo →Python • Bing API • Image Processing
FASHBuddy is a comprehensive fashion recommendation system that analyzes body shapes, fitness levels, and provides personalized fashion advice. The project includes image search using Bing API for fashion-related images, facial shape detection for neckline recommendations, shoulder and waistline detection for clothing style suggestions, and fitness comparison based on user metrics. It features modular scripts for analyzing different aspects of body morphology and generating tailored fashion recommendations.
Key Features: Facial shape detection, shoulder/waistline analysis, fitness comparison, fashion recommendations, Bing image search integration, and CSV output for merged results.
View GitHub Repo →HTML • JS • Educational
Learn Swedish in Stories is an educational web project that presents short, engaging narratives where vocabulary is taught in context. Each story includes inline translations, tooltip hints, and small interactive exercises to reinforce new words. The simple static site structure makes it easy to add new stories, adapt translations, and include multimedia assets. It's a low-friction way to build reading fluency and vocabulary retention through contextual exposure.
Educational design: scaffolded vocabulary lists, repeat exposure, and in-text tooltips that help learners pick up high-frequency words without breaking immersion.
View GitHub Repo →