Background: Diagnosis and classification of tumors is becoming increasingly dependent on biological and molecular biomarkers. RNA expression profiling using next generation sequencing (NGS) provides information on various biological and molecular changes in the cancer and in the microenvironment. We explored the potential of using targeted transcriptome and artificial intelligence (AI) in the differential diagnosis and classification of various hematologic and solid tumors. Methods: RNA from hematologic neoplasms (N = 2606) and solid tumors (N = 2038) as well as normal bone marrow and lymph node control (N = 806) were sequenced by NGS using a targeted 1408-gene panel. The hematologic neoplasms included 20 different subtypes. Solid tumors included 24 different subtypes. Machine learning is used for comparing two classes at a time. Geometric Mean Naïve Bayesian (GMNB) classifier is used to provide differential diagnosis across 45 diagnostic entities with assigned ranking. Results: Machine learning showed high accuracy in distinguishing between two diagnoses with AUC varied between 1 (Sarcoma vs GIST) and 0.841 (MDS vs normal control) (examples in Table). For differential diagnosis between all 45 different diagnoses, we used 3045 samples for training the GMNB algorithm and 1415 samples for testing. Correct first choice diagnosis was obtained in 100% of ALL, 88% of AML, 85% of DLBCL, 82% of colorectal cancer, 88% of lung cancer, 72% of CLL, and 72% of follicular lymphoma. The algorithm had difficulty in typically overlapping diagnoses and diagnosed as first choice 19% of MDS, 46% of normal, and 12% of MPN. Diagnosis improved significantly when second choice was considered. Conclusions: Targeted RNA profiling with proper AI can provide highly useful tools for the pathologic diagnosis and classification of various cancers. Additional information such as mutation profile and clinical information can improve these algorithms, reduce subjectivity, and minimize errors in pathologic diagnoses.