Agents that reason from represented knowledge rather than learned statistics — solving visual analogy problems and building a general game-playing agent.
Coursework on cognitive systems and knowledge-based approaches to AI — representing problems as structured knowledge and reasoning over that structure, rather than learning purely from data.
Built an agent that solves Raven's Progressive Matrices — visual analogy puzzles used in human IQ testing — by representing each figure as a set of objects and relationships, then generating and evaluating transformation rules to pick the answer that best continues the pattern. A second project extended this into a general game-playing agent capable of adapting the same representational approach across different game rule sets.
The agent matched average human performance across the standard Raven's test set, and the underlying representation transferred cleanly to the general game-playing task with minimal changes.