Arnav Mishra

I’m a results-driven ML Engineer focused on custom deep learning architectures, production deployment, and applied research. I aim to build reliable AI systems that bridge scientific insight with real-world impact.

In research, I developed Harmony-Aware Music Generation by enforcing music-theory constraints inside a two-stage Transformer using a custom Harmonic Relative Positional Encoding (H-RPE) layer and directional harmonic bias matrices, improving Self-Similarity Matrix Distance (published at IEEE CCIC 2025). I also built a hybrid quantum–classical deep learning framework for Gram-stained bacterial classification combining EfficientNetB0, PCA, and a 3-qubit variational quantum circuit—reaching 87% accuracy and 0.86 F1, outperforming both a classical CNN baseline and an initial hybrid model.

Currently, I’m working on Physics-Informed Longitudinal Progression Classification in Glioblastoma with a 3-stage pipeline (3D MRI segmentation, Fisher-KPP PINNs, and XGBoost) to distinguish pseudoprogression from true progression, including confidence-calibrated classification with human review escalation. My projects extend this mindset: NeuroBio generates falsifiable hypotheses grounded in biomedical evidence; SONAR 2.0 deploys a high-throughput geospatial ensemble to surface 135 potential archaeological sites; and my lung histopathology work fine-tunes ViT-Base to reach 99% accuracy with improved training efficiency and 100% reproducibility.