learning to lose slower or maybe not lose

[Uri Blass]

The idea is that you start with a losing position and play a game against deterministic engine

[d]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1
let say the opponent with black make a checkmate after 43 moves and first move was 1.e4

Now you try every different first move and play a game
If every different move does not help to lose slower then you keep 1.e4
If you find that 1.d4 lose after 44 moves and there is nothing better you learn to play 1.d4

You want to improve the loss after 1.d4 and you cannot do it by a different first move so you try to improve it by a different second move

1.d4 Nf6 2.Nf3 lost in 44 moves so you try every legal move that is not 2.Nf3 to find the slowest way to lose or maybe to draw or to win.
If you get a draw then you do not stop at the draw and try to improve every drawing move to see if you can get a win.

I wonder if there is some interface that can support this type of learning in order to find if there is a possible way to beat weak deterministic engines without a queen or maybe has more effective learning.

I wonder if there is some engine that can practically learn to lose slower or to draw or to win if you try many games against a deterministic engine.