← Work

Intelligent & Learning Agents

Bandit algorithms, MDP solvers, and a Minimax chess agent.

Period
August–November 2025
Kind
course
Stack
Python, Linear Programming
Guide
Prof. Shivaram Kalyanakrishnan, IIT Bombay

Three connected pieces of work on sequential decision-making.

  • Regret minimisation — implemented and benchmarked UCB, KL-UCB and Thompson Sampling, including a computationally optimised KL-UCB variant for stochastic environments.
  • Planning — modelled a strategic card game as a Markov Decision Process and solved it with Howard’s Policy Iteration and Linear Programming to derive optimal winning policies.
  • Search — built a Microchess agent using Minimax with custom heuristics, tuning search depth against execution time.