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
| Period | Institute | Project |
|---|---|---|
| Dec 2025 – Present | National Institute of Technology, Jamshedpur | Oral Cancer Detection · DWT-Gabor Swin Transformer |
| Aug – Nov 2025 | Indian Institute of Technology, Kharagpur | EEG · Autism Detection · P300 BCI |
| May – Jul 2025 | IIITDM Kurnool | Human Activity Recognition · BiLSTM-Transformer |
| Model | Task | My Result | Previous SOTA | Improvement |
|---|---|---|---|---|
| DCFNet (Thesis) | 3D Liver Segmentation · T2-MRI | 92.22% Dice | 86.51% | +5.71 pp |
| Inception-V3 + Huber | Fundus Image Quality (FIQA) | SRCC 0.9459 | TRIQ, HyperIQA | Top result |
| BiLSTM-Transformer | Human Activity Recognition | 95.80% Acc | BiLSTM baseline | Best of 6 |
| CNN | Lung Cancer · DICOM CT | 98.64% Acc | ANN 85.69% | +12.95 pp |
→ 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
| Project | What I built | Key metric |
|---|---|---|
| 🧬 DCFNet – 3D Liver Segmentation | Dual-encoder CNN · Cross-attention fusion · 3D MRI | 92.22% Dice |
| 🤖 AI-Interview-Assistant | LLM-powered technical recruiter · scoring dashboard | End-to-end AI pipeline |
| 🚂 RailSathiBE-Docker | Dockerized complaint system · Redis · JWT | Production-grade backend |
| 🔑 kpa-api | FastAPI · JWT auth · PostgreSQL REST API | Secure, scalable |
💬 I'm actively looking for PhD opportunities and research collaborations in AI-driven healthcare and computational pathology. Let's connect.
