Restaurant Billing & Ordering System Five role-based dashboards (Admin, Manager, Waiter, Chef, Cashier) backed by JWT + bcrypt auth and RESTful endpoints, live order/table state management via Redux Toolkit, Razorpay order creation with server-side payment signature verification, and an analytics layer aggregating sales data across 4 MongoDB collections and 20+ API routes. Tech stack: React, Redux Toolkit, React Query, Tailwind CSS, Node.js, Express, MongoDB, JWT, bcrypt, Razorpay. repo: https://github.com/whyatul/annmay-pos

Trust Verification System for AI-Generated Data A two-layer architecture where Layer 1 scores synthetic data on privacy (DCR, k-anonymity, membership-inference risk), utility (Wasserstein distance, ML efficacy vs. real data), and fairness (demographic parity, disparate impact), and Layer 2 anchors every result on Hyperledger Fabric via a multi-peer Byzantine fault-tolerant consensus engine. Smart contracts (chaincode) are implemented in Rust for performance, with a Dockerized multi-peer network (Hospital/Regulator/Research nodes) and automated GDPR/HIPAA/EU AI Act compliance checks. Benchmarked at under 5s verification latency for 10K rows with linear scaling to 50K. Tech stack: Python, Hyperledger Fabric, Rust (chaincode), CTGAN, Flask, Docker, Docker Compose. repo: https://github.com/whyatul/Blockchain-Based-Audit-Trail

Offline Voice Assistant An on-device pipeline (WebRTC VAD with a 300ms ring-buffer pre-capture window and adaptive noise-threshold calibration to eliminate clipped speech) feeding whisper.cpp for STT (5-10x faster than Python Whisper on ARM). A hybrid answer engine runs TF-IDF + fuzzy retrieval (RAG) first and falls back to a quantized TinyLlama 1.1B model served via Ollama for out-of-scope queries. A 3-tier cache (in-memory, disk, Redis) brings cached response latency under 1.5s versus 5-8s for a cold LLM call, and the whole stack runs as a systemd daemon for boot persistence. Tech stack: Python, whisper.cpp, Ollama (TinyLlama, Q4 quantized), Piper TTS, Redis, WebRTC VAD, systemd, Raspberry Pi. repo: https://github.com/whyatul/Voice_Activated_Chatbot

Digital Land Registry App Solidity smart contracts deployed via Truffle to the Polygon Amoy testnet (with a Ganache-based local network for offline development), a dual-party cryptographic authentication step required to finalize any ownership transfer, Mapbox-based geospatial coordinate capture for land parcels, and an image-upload flow for property documents alongside a live ETH-INR conversion API, all wired to MetaMask for wallet-based signing. Tech stack: Flutter, Dart, Solidity, Truffle, Polygon, Ganache, Alchemy RPC, Mapbox API, MetaMask. repo: https://github.com/whyatul/RegiTrust

Multi-User Chat Server A TCP server implemented directly on std::net with no networking framework, using a thread-per-connection concurrency model (std::thread::spawn per client) with a BufReader-backed blocking I/O loop for byte-level echo, plus a separate thread reading from stdin for server-side broadcast input. Tech stack: Rust, std::net, std::thread, BufReader. repo: https://github.com/whyatul/TCP-Server-Rust

Face-to-Face Video Calling App Dynamically generated call rooms (UUID-based) with an Express + EJS server rendering the signaling views, and a Socket.io channel handling SDP offer/answer and ICE candidate exchange to establish a direct WebRTC peer connection between the two clients. Tech stack: Node.js, Express, Socket.io, WebRTC, EJS, UUID. repo: https://github.com/whyatul/peer-to-peer-video-call

City Emissions Insights Dashboard A React frontend consuming a Flask REST API that serves three separate pre-trained scikit-learn models via joblib, a Random Forest regressor for emissions prediction, an anomaly-detection model to flag statistical outliers in the emissions data, and a time-series model for trend forecasting, over a CSV-based data pipeline with automated insight generation. Tech stack: React, Python, Flask, scikit-learn, joblib, Random Forest, time-series forecasting. repo: https://github.com/whyatul/SmartcityDashboard