Stephan Steinfurt is a senior consultant at TNG Technology Consulting and a physicist-turned-software engineer developing TNG AI Chess, which transforms real chess games into narrated instructional videos. His system addresses a fundamental mismatch: chess engines identify strong moves without explaining them, while language models explain ideas fluently but cannot reliably reason about the board unaided.
Steinfurt completed a doctoral dissertation in theoretical physics at Ludwig Maximilian University of Munich in 2014, with supervision at the Max Planck Institute for Physics. His research included fermionic diffusion in supersymmetric field theories and the hydrodynamics of spinning black D3-branes. By 2018, he was contributing to OpenADx autonomous-driving collaboration as a TNG participant in an industry workshop on simulation testbeds.
At TNG, he helped develop the TNG AI Chess YouTube channel, which converts Lichess games into educational videos through automated analysis, board annotations, generated narration, and publication. His approach centers on several practical engineering choices:
- Tool-grounded chess reasoning: Agents inspect legal moves, play and reverse variations, consult engine evaluations, and investigate checks, captures, and threats before explaining a position. A rook sacrifice becomes intelligible when the system shows how different responses expose the opponent’s queen to a knight fork.
- From scripted pipelines to autonomous agents: Early Python scripts assembled chess information for language models; later agents chose their own tools and explored candidate variations. Steinfurt tested Grok 4, OpenAI models, and Gemini 3.1 Pro against the practical demands of chess analysis.
- Instruction calibrated to human players: The system considers plausible human moves alongside optimal engine continuations, helping tailor explanations to different abilities. It also generates visual highlights, arrows, and ElevenLabs narration while balancing inference costs, publishing volume, and human review.
His AI Engineer Europe demonstration presented a clear ambition: making thoughtful, personalized chess commentary available for ordinary players’ games.
