const vraj: Human = {
education: "B.Tech Computer Science × MBA (dual-degree) @ Nirma University — IT",
leadership: "Technical Head, IEEE Student Branch NU",
focus: ["Computer Vision", "Federated Learning", "Applied ML", "Product"],
builds: ["full-stack apps", "vertical ERPs", "research prototypes", "hackathon wins"],
alsoDoes: ["music direction", "video & motion design", "teaching"],
philosophy: "learn by consequence, not by lecture",
status: "always shipping something",
};I live where research, product, and craft overlap — training vision transformers one week, shipping a QR-based ERP for marble yards the next, and scoring a stage play in between. If it can be built, benchmarked, or broken open to understand — I'm in.
mindmap
root((Vraj))
Research
Computer Vision
Foveal Vision Transformers
Road / scene extraction
Federated Learning
Healthcare benchmarking
Road-safety & EMS
Publication Integrity
Building
Vertical ERP · ShilaTeq
Hackathon prototypes
Full-stack web
Product
SMB & 0 to 1 thinking
Applied AI adoption
Roadmap strategy
Creative
Music direction
Video & motion
IEEE experiences
a live Mermaid mind-map — GitHub renders this natively, it's a real diagram, not an image.
|
IEEE Metaverse Grand Challenge 2026 · Smart Cities
A zero-backend, browser-based grid-crisis simulator. Learners manage a city's power grid through a Monte Carlo scenario engine, a real-time Learner Digital Twin, and a Gemini-powered Socratic AI advisor — grasping energy trade-offs through consequence.
|
Vertical ERP · commercialization in progress
A QR-based operating system for small Indian marble & granite yards — inventory, slab tracking, and yard ops digitized for a trade that still runs on paper. Built end-to-end and taken toward go-to-market.
|
|
A PyTorch training pipeline for a foveated ViT on ImageNet-1k — attention that mimics how the human eye samples detail. Deep-work into efficient vision architectures.
|
Privacy-preserving ML on two fronts: a healthcare benchmark on FLamby, and a road-safety / EMS framework on India's NCRB & MoRTH data — training without ever pooling sensitive data.
|
more in the arsenal — MethaneGuard (CV leak detection) · Truth-Anchor (vernacular fact-checking) · Predictive Maintenance (Azure IoT) · Publication Integrity (SPS) · Deep-Learning CNN / MNIST
| Languages | Python · TypeScript · JavaScript · C++ · PHP · LaTeX |
| ML & Data | PyTorch · TensorFlow · Keras · scikit-learn · NumPy · Pandas · SciPy · Matplotlib |
| Web & Build | React · Vite · Tailwind CSS · Node.js |
| Tools & Craft | Git · Jupyter · Premiere Pro · Canva |
beyond the code — click to expand
- Dual-degree B.Tech Computer Science + MBA at Nirma University's Institute of Technology
- Technical Head, IEEE Student Branch NU — building cinematic web experiences, orientations & challenges
- Builder of ShilaTeq / StoneX — a vertical ERP taking marble & granite yards from paper to product
- Music director for the Hindi stage play सत्ता / Satta — with video & motion design on the side
- Serial hackathon builder — Bharatiya Antariksh, HackOut, SEMICON India, SIH, UNESCO Youth & more
research & selected work — click to expand
| Area | Work |
|---|---|
| Vision Transformers | FoViT — foveated ViT training on ImageNet-1k |
| Federated Learning | Healthcare benchmarking on FLamby · Privacy-aware FL for road safety & EMS (NCRB/MoRTH) |
| Applied CV | MethaneGuard — methane-leak detection · road & scene extraction for urban mobility |
| Trust & Integrity | Multimodal scientific-publication integrity framework (SPS) · Truth-Anchor vernacular fact-checking |
| Industrial ML | Predictive maintenance on Azure IoT telemetry |
stay chaotic. ship anyway.
