Deep Learning Researcher · Medical Image Analysis · Healthcare AI · LLM Evaluator

Integrated M.Sc. Computer Science · Central University of Rajasthan, Ajmer, India (2021–2026)

I build deep learning systems that detect cancer from tissue slides, segment organs from 3D MRI, and decode brain signals — with results that beat published state-of-the-art benchmarks. I also design adversarial CS tasks that expose where frontier LLMs fail, bridging research and model evaluation.

📍 Ajmer, India  |  📧 hkp2857@gmail.com  |  LinkedIn  |  🎓 Seeking PhD / Research positions

→ llm-cs-eval-tasks — 15 real systems engineering tasks that expose LLM reasoning failures

I design and evaluate complex CS debugging tasks as part of active LLM benchmarking work. These tasks cover:

• Protocol bugs requiring RFC-level spec knowledge (DNSSEC, HTTP/2, UART)
• Systems debugging with no documentation — format must be reverse-engineered from binaries (WAL, eBPF, custom ISA)
• Multi-layer cascade failures where fixing layer A reveals bugs in B, C, D
• Custom codecs, embedded linker scripts, distributed tracing pipelines

Failure patterns I document: cascade reasoning, spec inference, simultaneous multi-layer repair, byte-identical precision, edge-case semantics

This work maps directly to: AI Evaluator · LLM Red Teamer · Benchmark Dataset Creator

🦷 Oral Cancer Detection — NIT Jamshedpur (Dec 2025 – Present)

• Designing a DWT-Gabor Fusion Swin Transformer for early-stage OSCC classification from H&E-stained histopathological images
• Novel 4th-channel input strategy: fusing multi-level Haar DWT + Gabor filter energy maps → stain-invariant frequency-domain features (cell boundaries, nuclear texture, tissue architecture)
• Replacing standard ViT patch merging with DWT-based patch merging — preserves edge and texture information critical for fine morphological distinction
• Benchmarking on NDB-UFES dataset (237 images, 3 classes) · correcting data leakage in prior work · 5-fold stratified image-level CV

🧬 DCFNet: 3D Liver Segmentation — Master's Thesis

• Dual-encoder architecture with Cross-Attention Fusion Module (CAFM): bidirectional cross-attention + learned per-voxel spatial gating
• Extended Coordinate Attention from 2D → 3D for volumetric medical image analysis
• Pipeline: PyTorch + MONAI · boundary-aware loss · deep supervision · mixed-precision · single NVIDIA T4 GPU
• Evaluated on 318 3D MRI volumes (CirrMRI600+) · 4 baselines · 6-variant ablation study · statistical significance testing

Deep Learning   Medical Image Analysis   3D Volumetric Segmentation   Computational Pathology
Oral Cancer Detection   Vision Transformers   Retinal Image Quality Assessment
EEG / BCI Systems   Wavelet-Based Feature Extraction   Healthcare AI   Time-Series Classification

ML / DL Frameworks  →  PyTorch TensorFlow MONAI Keras Scikit-learn
Medical Imaging  →  DICOM OpenCV OpenSlide Stain Normalization Patch Extraction Image Segmentation
Signal Processing  →  SciPy PyWavelets DWT Gabor Filters Band-pass Filtering Artifact Removal
Languages  →  Python C/C++ Java
Tools  →  Git Docker FastAPI PostgreSQL LaTeX Jupyter VS Code

💬 I'm actively looking for PhD opportunities and research collaborations in AI-driven healthcare and computational pathology. Let's connect.