In this work, we propose a new technique that reconstruct the Large Scale Structure using lyα forest in multiple sightlines from galaxies or QSOs. Based TensorFlow and able to utilize GPUs, this algorithm is 10-100 times faster than traditional method like Wiener Filter as implemented in Dachshund. We find the reconstructed error and accuracy of the cosmic web classification are both improved with our algorithm. We apply our algorithm to CLAMATO data and also predict the cosmic web reconstruction ability using next-generation telescopes, e.g., DESI and TMT.