Iron is an essential nutrient for most bacteria. However, iron limitation is a major barrier for pathogenic microbes due to host nutritional immunity, where iron is sequestered by human iron-containing proteins such as transferrin or in heme bound by proteins such as haemoglobin. This sequestration makes iron inaccessible to bacteria, and to infect the host, pathogenic bacteria must steal it using TonB-dependent transporters (TBDTs) present in the Gram-negative bacterial outer membrane. The TBDT ChuA from pathogenic Escherichia coli binds host haemoglobin in order to scavenge heme. In this study, we solved the structure of ChuA in complex with heme extracted from human haemoglobin, identifying residues required for coordination of heme. In addition, we modelled the ChuA-haemoglobin complex using AlphaFold and identified a hydrophobic haemoglobin binding region in the extracellular binding loops of ChuA, which is contiguous with the heme binding region. Based on these data, we have developed a putative mechanism defining initial ChuA-haemoglobin interaction and subsequent heme extraction. To test this model, we generated a panel of ChuA mutants in key residues from this region and validated their importance for binding haemoglobin and heme extraction using growth assays, and further purified them and characterised their ability to bind haemoglobin. Moreover, using RFDiffusion, we directed the design of 10,000 de novo proteins to bind at the ChuA:haemoglobin binding interface, and selected and tested 96 of the best candidates for their ability to bind ChuA and block growth on haemoglobin. We identified several binders that inhibit E. coli growth at low nM concentrations, without further optimisation. Finally, we determine the cryoEM structure of a subset of these binders, alone and in complex with ChuA, demonstrating that they closely match the computational design. As such this work demonstrates the utility of AI-designed de novo proteins in the design of next generation therapeutics.