Temple University Capstone Track

TeleGAssistant

An autonomous, privacy-first smart assistant that unifies university dashboards and communications into a single conversational interface.

View on Github

Project Vision

TeleGAssistant was designed to solve "dashboard fatigue" for modern students. Built by a collaborative 4-person software engineering team, the app connects directly to the Canvas API to monitor schedules, maps user context via SQLite, filters data through an on-device Large Language Model (Gemma 2 2B), and pushes smart updates straight to 'Telegram'.

By prioritizing edge computing and local AI inference over cloud APIs, the system guarantees absolute data privacy while maintaining a responsive, non-blocking notification ecosystem.

Core Ecosystem
đź“…

Smart Synthesis

Automatically aggregates Canvas data, emails, and calendars into a single operational point.

đź§ 

Privacy-First AI

Leverages a localized LLM to generate morning briefings without data leaving the machine.

⏰

Proactive Alerts

Monitors nearing task deadlines and updates tracking states to completely prevent user notification spam.

đź’¬

On-Demand Control

Features robust, interactive messaging commands allowing users to instantly request updates.

Module Distribution
Real-World Impact

đź”’ Eliminating Cloud Data Vulnerabilities

By routing Canvas academic records, private Gmail snippets, and local schedules strictly through a device-hosted Gemma 2 2B instance, the project introduces a highly secure alternative to traditional AI tools, keeping sensitive personal metrics completely decoupled from third-party vendor servers.

⚡ Resolving Student Dashboard Fatigue

The application condenses fractured academic environments—including course updates, due dates, and grading rubrics—into a singular conversational feed. This design actively prevents platform hopping and ensures critical milestones are delivered as prioritized summaries directly to a user's phone.

🛠️ Scalable Open-Source Architecture

The system demonstrates a robust blueprint for local-first utility middleware. Leveraging an asynchronous, event-driven monolith model running on Python and SQLite proves that powerful, personalized automation apps can be reliably sustained on low-overhead consumer hardware without cloud subscription models.

đź’ˇ Accessible, Conversational Data Science

TeleGAssistant simplifies structural data parsing, API synchronization, and prompt engineering into an intuitive text-based user workflow. It bridges complex technical pipelines with conversational UX design, demonstrating how advanced AI engineering can be translated into straightforward, everyday software value.