milan = {
"institution" : "NIT Surat — B.Tech EE (CGPA: 9.02)",
"focus" : ["ML-DL Engineer","Data Science","Computer Vision","ML-DL Resaercher"],
"papers" : ["CVPRW 2026 (MELD-SR)", "Springer CCIS — NCVPRIPG 2025 (PRISM)"],
"role" : "Research Head @ Sillycon AI Labs",
"competing" : "NTIRE 2025 & 2026 — Top-20 finishes across 4 tracks",
"status" : "🔍 Hunting for real-world chaos to throw models at",
}| 🥇 Achievement | Detail |
|---|---|
| 📄 CVPRW 2026 | MELD-SR — Multi-Expert Frequency-Guided SR ×4 (Rank 19 / 17) |
| 📄 Springer CCIS | PRISM — Multi-attention Transformer for SR, NCVPRIPG 2025 |
| 🏅 NTIRE 2026 | Lumi-Drop (Rank 12), WREN (Rank 13), MELD-SR (Rank 17/19) |
| 🏅 NTIRE 2025 | HAT-L based SR ×4 — Rank 9 Restoration & Perceptual |
| 🔬 Research Intern | IIT Indore — Novel hyperspectral reconstruction, outperforming MST++ |
| 👥 Research Head | Leading 15-member AI/DL research lab at NIT Surat |
Languages
AI / DL Frameworks
Architectures I work with
MLOps & Deployment
| Project | Description | Result |
|---|---|---|
| 🌧️ Lumi-Drop | Restormer + Day/Night NAFNet experts with dynamic luminance routing for raindrop removal | PSNR 27.57 dB · NTIRE'26 Rank 12 |
| 📐 MELD-SR | Frequency-guided (DCT/DWT/FFT) multi-expert SR fusion — published at CVPRW 2026 | PSNR 31.48 dB · NTIRE'26 Rank 17/19 |
| 🔊 WREN | Heterogeneous ensemble denoising (Restormer + NAFNet) with learnable softmax mixing | PSNR 28.99 dB · NTIRE'26 Rank 13 |
| 🔋 Battery Anomaly Detection | End-to-end LSTM Autoencoder + FastAPI/Streamlit + CI/CD on AWS EC2 | Production MLOps system |
| 🧠 DL Implementations | PyTorch implementations of neural nets on standard benchmarks | Learning resource |
- MELD-SR: Multi-Expert Learned Decomposition Using Frequency Guidance — CVPRW 2026
- PRISM: Progressive Regional Integration with Synergetic Multi-attention Transformer for SISR — Springer CCIS, NCVPRIPG 2025
