Advances in multimodal single cell analysis can empower high-resolution dissection of human vaccination responses. The resulting data capture multiple layers of biological variations, including molecular and cellular states, vaccine formulations, inter- and intra-subject differences, and responses unfolding over time. Transforming such data into biological insight remains a major challenge. Here we present a systematic framework applied to multimodal single cell data obtained before and after influenza vaccination without adjuvants or pandemic H5N1 vaccination with the AS03 adjuvant. Our approach pinpoints responses shared across or unique to specific cell types and identifies adjuvant specific signatures, including pro-survival transcriptional states in B lymphocytes that emerged one day after vaccination. We also reveal that high antibody responders to the unadjuvanted vaccine have a distinct baseline involving a rewired network of cell type specific transcriptional states. Remarkably, the status of certain innate immune cells in this network in high responders of the unadjuvanted vaccine appear "naturally adjuvanted": they resemble phenotypes induced early in the same cells only by vaccination with AS03. Furthermore, these cell subsets have elevated frequency in the blood at baseline and increased cell-intrinsic phospho-signaling responses after LPS stimulation ex vivo in high compared to low responders. Our findings identify how variation in the status of multiple immune cell types at baseline may drive robust differences in innate and adaptive responses to vaccination and thus open new avenues for vaccine development and immune response engineering in humans.