I could be your developper, but I would be paid for it 😄
For the examples you show, seems more that what you are looking for is just a trigger, as in an oscilloscope. So a trigger is easy to implement, as it is a simple condition: (sample[t-1]< Trigg) &&(sample[t] > Trigg)
You can even use multiple triggers to detect falling edges and rising edges with different trigger level. Then, if you need more accuracy in the peaks, you could use peak detection but only after the trigger fired, or, you can just take the middle point between a falling and a rising edge.
Sorry for the hand-drawed picture, but this would be the final times you would detect if you use triggers and middle points. I would choose middle points because it seems that your signal is "saturating" at some point, having flat "plateau" instead of a peak, which will lead to some difficulties if you try peak detection.
