On-demand access to information encoded in nucleotides lies at the heart of DNA/RNA applications. However, contemporary methods for targeted retrieval using PCR amplification or bead-based extraction, rely on Watson-Crick base pairing and pre-defined primers, limiting dynamic decision-making whilst sequencing. We introduce SUSTag-ORCtrL, a nanopore-based system enables real-time, PCR-free random access to DNA-stored data by directly classifying raw ionic current signatures of 96 or 384-plex DNA molecular tags. Our framework combines SUSTag, a Bhattacharyya distance and incremental clustering enhanced molecular tag design (SUSTag) to minimize crosstalk, with an Optional-Reject Cnn-lstm deep learning model inspired by TRansfer-Learning (ORCtrL), designed to enhance adaptability to signal variability. SUSTag-ORCtrL achieves an intra-class weighted F1-score of 99.69% for 96-plex classification and 99.05% for 384-plex classification, surpassing existing molecular tagging systems. Domain adaptation using only120 minutes of new sequencing data ( ~ 200k reads) boosts the model performance from 87% to over 94%, achieving complete recovery of target data with minimal crosstalk in 10 min to 3 h. This system provides a scalable, low-latency solution for versatile DNA data access and holds promise for genomic and transcriptomic disease screening and the DNA-of-things.