🌍 NIT Rourkela, India · 🎯 ML → production web apps · 🔧 React, Three.js, Supabase
🛠️ opencode · hermes-agent · cursor · claude · ⚡ learn by shipping, iterate fast
💼 Open for: ML / full-stack internship opportunities
I build things that matter. Right now that means splitting my energy between machine learning fundamentals (linear regression → neural nets) and modern full-stack web development (React, TypeScript, Three.js, Supabase). My goal is to bridge the gap — ship ML-powered web apps end to end, from trained model to deployed product.
Real-time sentiment analysis and signal generation engine. Aggregates Reddit, news, and market data into actionable Fear & Greed scores and trade signals via Apache Kafka.
├── 🧠 FinBERT ONNX for sub-100ms sentiment analysis
├── 📡 Kafka pipeline: producers → consumers → signals
├── 📊 Fear & Greed Index (sentiment 50% + momentum 30% + velocity 20%)
├── 🎯 Contrarian & trend-following signal strategies
├── 🐍 Python 3.11+ with Poetry
└── 🐳 Docker Compose for Kafka + Zookeeper
Stack: Python · Kafka · FinBERT · Docker · Pydantic · YAML Config
End-to-end ML system for detecting fraudulent credit card transactions. Three classifiers (Logistic Regression, Random Forest, XGBoost) with SMOTE for class imbalance, a FastAPI backend, and a live React dashboard.
├── 🤖 Three ML models with threshold-tuned predictions
├── 📊 SMOTE oversampling for 0.17% fraud class
├── ⚡ FastAPI REST API with /predict, /predict/batch, /models
├── 📈 React 19 + Recharts live dashboard
├── 🔬 MLflow experiment tracking
└── 📓 Jupyter notebooks for EDA & training
Stack: Python · Scikit-learn · XGBoost · FastAPI · React 19 · TypeScript · MLflow
The official website for NIT Rourkela's premier aeromodelling club. A production-grade platform for team induction, event registration, and member management.
├── 🎨 3D interactive hero powered by Three.js
├── 📝 Event registration with real-time Supabase backend
├── 🔐 Member authentication & role-based access control
├── ⚙️ Admin toggle panel for induction & registration windows
├── 📊 Excel/CSV export for applicant data management
├── 📱 Fully responsive across all device sizes
└── 🚀 Deployed on Vercel with CI/CD
Stack: React · TypeScript · Three.js · Supabase · Framer Motion · Vite · Firebase
| Area | Metric | Value | Details |
|---|---|---|---|
| ML | Fraud detection | 97.5% ROC-AUC | XGBoost with SMOTE + threshold tuning |
| ML | Sentiment latency | <100ms | FinBERT ONNX, batch of 32 |
| 3D | Render performance | 60 FPS | Three.js hero scene optimized draw calls |
| 3D | WebGL draw calls | < 50 | UDAAN interactive scene |
| Backend | Query latency | < 100ms | Supabase real-time queries |
| Streaming | Throughput | 1000+ msg/min | Kafka pipeline across all topics |
| UX | Breakpoints | 5 | Mobile-first responsive design |
| Auth | Role management | Admin / Member | Supabase RLS policies |
▓▓▓▓▓▓▓▓░░░░░░░░░░ Fundamentals [██████████░░] 65% — supervised, ensembles, SMOTE, evaluation
▓▓▓▓▓░░░░░░░░░░░░░ Deep Learning [██████░░░░░░] 35% — NLP, transformers, ONNX inference
▓▓▓░░░░░░░░░░░░░░░ Advanced [███░░░░░░░░░] 15% — real-time pipelines, model optimization
Built: Classification (LR, RF, XGBoost), SMOTE, NLP sentiment (FinBERT), model serving APIs Next: Training transformers from scratch, CNNs, deployment at scale
□ Master ML fundamentals — statistics, linear algebra, supervised learning
□ Build an ML-powered web app — model training → API → frontend integration
□ Contribute to open-source projects
□ Ship 2+ portfolio projects this year
□ Land an ML or full-stack internship
Learning in public. Building in public. Growing in public.