AI Architect – Tech Lead, Mahjong Game

This is a fully remote position, open to applicants in Poland.

📋 Description

• Design and develop the algorithm for playing mahjong utilizing imitation learning, reinforcement learning, search with MCTS, or a combination of these methods.

• Train, assess, and deploy the mahjong-playing model.

• Take ownership of the complete technical architecture, which includes the game model, LLM reasoning layer, valid-action mask, win detection, and the API contract with the game bridge.

• Design and evaluate the inference path, batching, and caching to ensure compliance with the 2-second response time requirement.

• Specify the necessary client data and event names, such as hand histories and event streams.

• Create the training and calibration pipeline utilizing client data.

• Develop and operate an evaluation harness for play strength and explanation quality.

• Implement LLM observability with Langfuse using asynchronous logging and N+1 batching.

• Transition context from Vlad Borysenko during the ramp-up phase and lead sprint efforts alongside the AI Engineer.

• Collaborate with the client's Product Owner within a scrum framework.

• Communicate technical decisions and trade-offs to the client's CTO and engineering team.

• Track risks associated with licensing new training data, engine-bridge functionalities, and the scope of multi-rule sets.

⛳️ Requirements

• A minimum of 6 years of hands-on experience in ML/AI engineering, specifically in real game AI or sequential decision-making projects.

• Practical experience with RL, imitation learning, or search-based agents like MCTS and self-play.

• Familiarity with imperfect-information games, ideally including mahjong, poker, or other card games.

• Proficiency in Python for ML system development.

• Strong software engineering capabilities, encompassing APIs, testing, and CI practices.

• Comprehensive experience in end-to-end model training using user or gameplay data: data → training → evaluation → serving.

• Expertise in LLM application engineering, including reasoning layers, prompt and context design, structured outputs, and guardrails.

• Experience in low-latency inference, including profiling, batching, caching, and managing model-size trade-offs.

• Knowledge of LLM observability and evaluation techniques with Langfuse or similar tools.

• Experience with AWS deployment for ML workloads.

• Understanding of game theory as it applies to imperfect-information games.

• Ability to evaluate play strength through win rates, Elo-style ratings, and baseline agents.

• Familiarity with game-engine integration patterns, including event streams, action masks, and state bridges.

• Knowledge of AWS Well-Architected principles for ML workloads.

• Proven hands-on experience in ML/AI engineering at a production scale.

• Experience in deploying an AI system within a live product that has stringent latency constraints.

• Background in cloud hyperscalers, preferably AWS.

• Experience in technology consulting or client-facing delivery roles.

• Proven ability to train models on gameplay data from start to finish.

• Experience leading small delivery teams while also contributing code personally.

• Capacity to manage an architecture in front of a technical client CTO.

• Advanced proficiency in English is required.

• Availability for part-time remote work.

• Willingness to collaborate as a B2B contractor.

🏝️ Benefits

• Part-time engagement at 0.5 FTE for a duration of 3 months.

• Opportunity for remote work.

• Collaboration in a B2B (contractor) format.