← Back to Portfolio
Personal · project

Alakazam-VGC: Natural Language Pokémon Battle Calculator

LLM-powered web application enabling competitive players to perform complex game calculations through conversational interfaces

A specialized web application that transforms traditional Pokémon VGC calculations into natural language interactions using fine-tuned large language models. The system replaces complex manual calculations with conversational queries, enabling competitive players to perform speed checks, EV optimization, and stat analysis through intuitive chat-based commands.

Key Features

Multi-Intent Processing: Handles three distinct calculation types - speed comparisons, optimal EV training recommendations, and base stat queries • Fine-tuned LLM Integration: Custom language model specifically trained for Pokémon VGC terminology and calculation patterns • Performance-Optimized Deployment: Streamlit Cloud implementation with intelligent caching system reducing response times from 15 seconds to 4 seconds • Session Persistence: Built-in calculation history tracking allowing users to reference previous results within the application

Technical Implementation

The application demonstrates sophisticated NLP pipeline architecture, combining fine-tuned language models with domain-specific parsing logic to interpret competitive gaming terminology. The implementation showcases modern ML deployment patterns through Streamlit's cloud infrastructure, with intelligent caching strategies that optimize user experience by learning from interaction patterns.

Skills & Technologies

Large Language Models Natural Language Processing Streamlit Python Machine Learning Fine-tuning Web Development Caching Command Line Interface
// stay in touch

Get Answering Machines in your inbox.

Occasional notes on AI, coding agents, and where this is all headed — plus the very occasional origami diagram.

Arjun Kirti Patel
Chicago, IL · folded with care
© 2026 Arjun Kirti Patel · MIT Licensed