We demonstrate a prototype of a multi-algorithm learning and classification engine on a state-of-the-art FPGA. The architecture can accelerate training and classification for CNN, SVM and GVLQ. It can also do Supervised Semantic Indexing (SSI). We identify common underlying computations in these various algorithms. Specifically, we map them to variants of matrix-matrix and matrix-vector operations, and architect a many-core coprocessor to execute a massively parallelized version of these operations using fine-grained, lightweight threads. A 512-core prototype will be demonstrated on a Xilinx SX240T FPGA on an off-the-shelf PCI card.