We present here MADaM (Multiplexed Amplicon Data Miner), an original algorithm designed to de-novo genotyping of small sequencing reads that do not require assembly step. It performs a classification of the reads based on an original set of features using t-SNE’s and clustering with the DBSCAN algorithm. The algorithm is applied to three different approaches and datasets showing that this software is fully suitable for fastly genotyping highly variable regions such as MHC-HLA exons 2 without any priors such as SNP positions or already known alleles.