We developed a multi-instance deep learning framework called EAMil to diagnose systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA) with high accuracy using TCR sequencing data. By integrating PrimeSeq feature extraction, ESMonehot encoding, and an improved gated attention mechanism, we achieved state-of-the-art performance with AUCs of 98.95% for SLE and 97.76% for RA. EAMil successfully identified disease-related genes with over 90% agreement with conventional differential analysis and effectively distinguished disease-specific TCR genes. We demonstrated robustness in classifying SLE patients according to disease severity and lesion site diagnosis using the SLEDAI score, effectively controlling for confounding factors such as age and gender. This interpretable immune receptor analysis framework provides new insights into autoimmune disease detection and classification with broad potential clinical applications across the spectrum of immune-mediated diseases.