Optical biosensors offer significant advantages for minimally invasive in vivo and in vitro diagnostics. Optical sensors that are based on the fluorescence of semiconducting single-walled carbon nanotubes (SWCNTs) especially benefit from a near-infrared (nIR) signal that is sensitive to chemical perturbations and overlaps with the tissue transparency region where the absorbance of water and blood is minimal. Unlike traditional fluorophores, SWCNTs do not photobleach, making them ideal for long-term measurements in untreated samples. Single-stranded DNA (ssDNA) sequences are natural polymers that self-assemble on the SWCNT surface through pi-pi stacking. This DNA wrapping solubilizes the nanotubes in aqueous solutions and modulates the selectivity and sensitivity of the sensor towards a specific analyte. The properties of ssDNA-wrapped SWCNTs (ssDNA-SWCNTs) depend on the DNA base composition and length. Despite significant advancements in ssDNA-SWCNT biosensors, the sensing mechanism remains not fully understood. In this thesis, we explore systematic approaches for engineering ssDNA-SWCNTs for cancer detection. We investigate a spectrum of engineering methods, from a guided semi-rational design to more empirical approaches, based on the extent of understanding of interactions between the ssDNA-SWCNTs and analyte. We first aim to develop ssDNA-SWCNTs for sensing the redox-active molecule glutathione (GSH) under varying pH conditions. We observe a substantial pH effect that changes with the DNA sequence and exploit this pH sensitivity for applications in cancer detection. By employing semi-rational engineering with xeno nucleic acids (XNAs), we further engineer optical biosensors that enable the detection of GSH in the presence of contributions from pH. We also performed additional characterizations to elucidate the sensing mechanism and to investigate the thermodynamics of the interactions. In the next chapter, we design ssDNA-SWCNT biosensors for serum biomarkers associated with Pancreatic Ductal Adenocarcinoma (PDAC). In the absence of information on DNA-biomarker interactions, we employ an irrational design approach, directed evolution. Additionally, we investigate the potential of machine learning assistance through a novel DNA clustering approach to systematically explore the vast DNA sequence space and enhance the directed evolution process. In the following chapter, we investigate the detection of biomarkers in complex media. We utilize a diverse DNA library designed through machine learning to form a sensor array. We monitor multiple SWCNT chiralities to develop an optical fingerprint for the biomarker in the presence of an interfering protein. The strategies developed in this thesis tackle longstanding bottlenecks in the field of creating ssDNA-SWCNT sensors. They allow us to design sensors with little to no information on the mechanism. Importantly, they uncover novel solutions for sensing biomarkers in complex media for practical applications.