Chitti — AI That Catches What Matters

Students live in noisy group chats where important information is constantly buried under messages, forwards and conversations. A submission deadline, fee reminder, timetable change or request can disappear within minutes. The problem is not that students lack calendar or to-do applications—the missing step is capturing important information at the moment it arrives. Chitti addresses this gap by acting as a personal intelligence layer on the Android phone. It continuously processes incoming notifications, removes duplicates, applies a lightweight keyword/date filter and sends only potentially important messages to an on-device AI model. The model identifies actionable commitments such as deadlines, payments, meetings and requests, extracts useful details such as what, when and who, and converts them into one-tap action cards. For example, when a student receives “Record submission tomorrow 10 AM, Lab 2,” Chitti can identify the submission deadline and present an action to add it to the calendar or create a reminder. A Telugu-English message such as “25th lopu fee kattali, marchipoddu” can also be interpreted as a fee-payment commitment. This multilingual, conversational understanding reflects how students actually communicate in WhatsApp and other notification-driven environments. The target users are students and other smartphone users who receive high volumes of notifications and frequently depend on group communication for deadlines, payments, meetings and requests. Users interact through a Today screen, voice or text-based Ask interface, reminders, calendar actions and a personal Library of captured information. Privacy is a core part of the architecture, not an optional feature. Notification processing and language-model inference are designed to happen on the device. The application can continue working without network connectivity for its core intelligence, supporting a phone-first experience even in airplane mode. The architecture uses Android NotificationListenerService, SHA-256-based de-duplication, keyword/date filtering, Gemma 2B through MediaPipe with a rule-based fallback, Room for local storage, and Android alarms, calendar and sharing capabilities. Chitti also extends this intelligence to safety and personal productivity. LinkGuard checks potentially dangerous links before opening them, while the application includes voice interaction, personal memory, document processing and secure autofill capabilities. What makes Chitti distinctive is the point where it intervenes: before information becomes a forgotten task. Instead of asking users to manually copy messages into another productivity application, Chitti captures intent at the notification layer and turns it into an actionable outcome. The result is a personal, contextual and privacy-conscious assistant designed around how people already communicate.