Lowkey Warrior

Our AI agent is designed to strategically outperform opponents in Negotiation, Poker, and Spelling Bee by leveraging a custom Model-Context-Protocol (MCP) framework, retrieval-augmented generation (RAG), and Supabase for structured game intelligence.

Key Technical Contributions
Supabase for Game Data Storage & Retrieval

Stores past game logs, including move history, outcomes, and reasoning (info[“reason”]) to refine future strategies.
Enables persistent learning, allowing the AI to adapt its gameplay over time based on previous wins and losses.
Retrieval-Augmented Generation (RAG) for In-Game Strategy

Rules, strategic patterns, and past game moves are indexed in a vector database for real-time retrieval.
During each game turn, the AI queries relevant past experiences and retrieves optimized strategies to make informed decisions.
Custom MCP for Spelling Bee Optimization

Dictionary Initialization: Loads a comprehensive word dictionary into memory.
Dynamic Word Filtering: Finds all possible words using the given set of letters.
Optimized Word Selection: Returns words sorted by length, prioritizing the longest possible word to ensure optimal performance and guaranteed wins when playing first.
Deception & Bluffing Strategies for Negotiation & Poker

Negotiation: The agent misrepresents resource valuations and strategically delays acceptance or counter-offers to maximize its trade benefits.
Poker: Implements bluffing patterns, varying bet sizes to manipulate opponent behavior while referencing statistical game theory models for optimal betting.