This project pairs a Python modeling pipeline with a small React front end. The backend cleans MyAnimeList data, turns synopses into TF-IDF features, combines those with metadata, and trains multiple models for score prediction.

I treated it as a practical experimentation environment, so the work emphasized leakage-safe validation, useful feature engineering, and repeatable model comparison rather than novelty for its own sake.

The frontend gives the models a cleaner face with an anime card grid, score badges, and CSV export so the results are easier to inspect and share.

CraveAI is a conversational Retrieval-Augmented Generation chatbot built to recommend restaurants based on how someone feels, what they are craving, and where they are in real time. It combines LangChain, the OpenAI API, Google Places, and vector retrieval to turn a casual prompt into a useful local recommendation.

I implemented natural-language intent parsing, vector-based cuisine retrieval with ChromaDB, and an interactive React plus FastAPI experience that keeps the full recommendation flow fast and readable for users.

The result was a sub-second recommendation interface with strong engagement, plus a cleaner product framing for restaurant discovery than a standard search filter UI.

This site started as a long list of project cards. The redesign changes it into a curated archive where visitors scan the left rail and read one project at a time in a more deliberate detail view.

That shift brings the presentation much closer to the reference image you shared while keeping the site lightweight and fully static.

XY-Ball-Fight is a lightweight 2D arena game built without a framework. The focus is on fundamentals: a game loop, collision handling, controls, weapons, and readable UI systems.

• Vanilla JavaScript implementation with no build step.
• Custom physics and collision logic.
• AI opponents and modular engine files.

Chess-Bot explores classic search-based AI rather than machine learning. The core challenge is evaluating board states efficiently enough to make useful decisions in a reasonable time budget.

It is a good example of adversarial search, state representation, and performance-aware implementation.

Evodle mixes incremental game structure with simulation ideas. Instead of progression being purely linear, the system uses evolutionary mechanics to produce more interesting upgrade behavior over time.

This package bundles together reusable NLP workflows for classification, keyword extraction, and sentiment analysis so the same preprocessing and language steps do not need to be rebuilt from scratch each time.

VanklCommApp focuses on live communication on mobile, with chat, group channels, and file sharing supported by Firebase. It highlights product work where state synchronization and UX both matter.

ACMERun tracks runs, pace, and heart rate while syncing health data with Google Fit. It is a good example of mobile product work that mixes sensors, data, and interface design.

Island Generator explores terrain synthesis through Perlin noise and erosion-inspired post-processing. The goal was to make generated landforms feel more organic than raw noise alone.

This library turns terrain data into reusable 3D geometry with UV mapping and configurable shader support, making it more of a graphics building block than a one-off demo.

Piraten-Karpen blends exploration, procedural map generation, inventory management, and a playful retro visual style. It captures the side of the portfolio that leans into game systems and presentation together.