Abhishek Pal

I’m an Artificial Intelligence Consultant building agentic medical imaging platforms end-to-end. I designed “BraTS-Agents,” an autonomous multi-agent pipeline that adapts preprocessing and tensor schemas to each dataset’s DICOM/NIfTI format, modality availability, and voxel spacing—without hard-coded rules.

My work combines LLM-driven code generation with deep learning research for runtime adaptability. I implemented the preprocessing backbone using ANTs rigid+affine longitudinal registration, 1mm isotropic resampling, SegResNet tumor segmentation, blank-slice removal, and 8-channel PINO tensor construction with Fourier time features, containerizing the platform with Docker and tracking experiments via W&B across Azure ML and AWS H100 GPU clusters.

Earlier, as an Artificial Intelligence Intern, I implemented Physics-Informed Neural Operators (PINO) and Fourier Neural Operators (FNO) under PDE constraints (Fisher-KPP / reaction-diffusion), trained on NVIDIA H100 GPUs, and improved accuracy by 20% using Stable Diffusion–based synthetic augmentation. I also ship applied work—NeuralTraffic with YOLOv8 (52%→78%) and a dispatch strategy reducing congestion by 20%, plus DraupadiAI with Librosa MFCCs and a 92% accurate distress detection pipeline paired with on-device LLaMA 7B for privacy-first counselling.