A collection of research, technical implementations, and experiments in AI and Software Engineering.
Integrated genomic, transcriptomic, and proteomic data to identify predictive biomarkers of liver cancer progression. Applied deep learning and statistical modeling methods to complex multi-omics data pipelines.
Developed methods to quantify the effects of domain shift on RL agents using reactive exploration. Implemented experiments across multiple OpenAI Gym environments to evaluate predictive capability.
Built an end-to-end pipeline from TCGA mutations to peptide candidates. Modeled peptide–MHC binding with GNNs trained on IEDB data and validated with UniProt context to rank neoantigens.
Implemented a 3D MixTransformer backbone with overlapping 3D patch embeddings and stochastic depth. Produced voxel-wise masks via trilinear upsampling for volumetric anatomy.
Tokenized 2D slices into patch sequences and applied multi-head self-attention. Aggregated slice-level CLS tokens using dedicated inter-slice attention for volume-aware diagnosis.
Analyzed GRN outputs to derive network statistics and consensus interactions. Integrated ranked interactions into downstream survival/response feature sets.
Implemented agents under domain shifts, including spaced-repetition-style training to test recall robustness across classic control and Atari tasks.
Reproducible NixOS system configuration using flakes. Includes Hyprland, development environments, and container tooling.