Maram Maruthi Chethan

CalmSend – AI Emotional Writing Copilot

• Built a Manifest V3 Chrome extension injecting a live panel into WhatsApp Web, Instagram, Gmail,and Google Docs — intercepts compose boxes before send, Grammarly-style. • Designed a hybrid NLP ensemble: word TF-IDF + char TF-IDF blended at 55/45(tuned via grid search on validation F1); trained on 17,280 samples and validated at 95% accuracy on 3,802 test samples across safe, caution, and high-risk classes. • Implemented app-aware cooldown logic to handle sarcasm, profanity, and aggressive tone; delay adapts per app, content type, and model confidence; dynamically blocks Send action. • Built FastAPI backend with live model retraining without server restart. Won 1st Place at ProjectExpo

SetuAI – Government Data Query System

• Built an NL query console across 3 departments (Agriculture, Education, Urban Housing) with OpenAI API + heuristic fallback, guaranteeing uptime independent of external APIs. • Implemented HMAC-SHA256 JWT authentication with timing-attack-safe verification and role-based access control (RBAC) — each department restricted to its own data. • Improved query reliability by handling ambiguous inputs through intelligent fallback to full department datasets. • Deployed via GitHub Actions CI/CD to GitHub Pages; all queries logged with timestamp, role, and result count for audit and scalability analysis.