 Today I present our closed-loop non-invasive neuroprosthetic research platform where we connect a living biological neural network to an artificial hand outfit with a tactile sensor. This process begins by selecting an efferent extracellular recording site, decoding the information, and initiating the tap of a finger. Once the tactile sensor comes into contact with the environment, this information is encoded where we then stimulate back into an afferent extracellular stimulation site. Neural decoding takes multi-unit action potentials, and when they exceed a certain threshold of number of spikes per second, outputs a movement to the joint controller for the hand. The sensory encoding allows that tactile experience to be able to be encoded into slowly adapting and rapidly adapting mechanoreceptor firing patterns which are stimulated into the culture. This then completes our closed-loop non-invasive neuroprosthetic platform. Our system level diagram begins at our head stage where we have an efferent recording site, we detect spikes based on the extracellular voltage, we then sum those spikes within a bin to produce a 100 millisecond pulse called MEA out, which we relay to our teensie, which is Rossnode 1. Rossnode 1 communicates with the Rossmaster, where Rossmaster uses a joint controller to output a no-contact state or a contact state depending upon MEA out. This information produces a tactile experience, resulting in a force of DC and AC components, which are encoded utilizing an Izikovich neuron model. This information is relayed through an action potential generator board, where that information is sent to stimulate the MCS SCU, which then relays that information to a stimulation site within the MEA culture. We use extracellular recording information to extract spikes. Those spikes are utilized to control the joint angle of the robotic finger. This brings the finger in contact with the environment to produce FDC and FAC components of pressure. Those are encoded into mechanoreceptor firing patterns which are stimulated back into our afferent stimulation site. Offline analysis is performed by taking the extracellular recording information in C and using a 512-point FFT to produce the information in B. We then utilize the spike information in D and utilize the same 512-point FFT to produce the information in A. We extracted windows of time to train neural networks. The artificial neural network classification accuracies are shown for cultures 1 and 2. Additionally, we looked at intertap intervals for culture 1 on DIV35 and culture 2 on DIV20-23 across RA and SA with statistical significance for each culture.