Added Mikhail Parakhin as an advisor and mentor to the project.

Production deployment of LFMs achieves sub-20ms inference on real workloads with approximately 50% fewer parameters and 2-10× faster performance than alternatives.