Project Description
Senior Project Description
The internet is a vast resource of information, yet not all web content is easily consumable by every user. This project focuses on developing an intelligent on-device web assistant that performs user-specific renarration of web content, automatically rearticulating information in ways that align with different user needs, contexts, and capabilities. Rather than simply providing content, this assistant serves as a sophisticated intermediary that helps users navigate, understand, and effectively utilize web resources. The aim is to make digital information more inclusive and adaptable, ensuring that content is not just available but truly understood by diverse audiences while preserving complete user privacy through local processing.
Renarration and Accessibility
Renarration is the process of transforming content to match the perspective, background, or situational needs of a given audience. This project positions the system as an intelligent assistant that understands both the user and the content, facilitating more effective web utilization. Accessibility in this context extends beyond traditional disabilities to include anyone who may struggle to engage with web content due to:
- Vision impairments, making it difficult to consume text-based content
- Educational background, where complex or technical language creates barriers
- Language proficiency, where direct translation alone is insufficient for comprehension
- Contextual needs, such as requiring concise summaries while multitasking or preferring in-depth analysis
- Cultural differences, where content assumes specific cultural knowledge or references
For example, a news article about a historical event could be renarrated differently for:
- An aged grandparent, emphasizing personal and societal impact with connections to their lived experience
- A younger sibling, simplifying key facts and using familiar comparisons from their generation
- A retired history professor, providing deeper academic context and historiographical perspectives
- A non-native speaker, adjusting cultural references and providing the necessary background context
Technical Approach: On-Device Intelligence
The system will leverage state-of-the-art AI models deployed entirely on user devices, addressing critical technical and privacy challenges:
On-Device Processing Advantages
Privacy Preservation: All user interactions, browsing patterns, personal preferences, and renarration requests remain completely local. No sensitive data is transmitted to external servers, ensuring absolute privacy protection—particularly crucial when users access medical information, personal research, or sensitive topics.
Enhanced Accessibility: On-device processing eliminates internet dependency for core renarration functionality, providing consistent assistance even with poor connectivity. This democratizes access for users in regions with limited bandwidth, those on restrictive data plans, or in situations where connectivity is unreliable.
Reduced Latency: Local processing enables real-time content transformation without network delays, creating a seamless browsing experience where renarration happens instantly as users navigate web pages.
Core Technical Challenges
Model Optimization: Implementing efficient model compression and quantization techniques to fit capable Language Models and Vision-Language Models within mobile device memory and processing constraints while maintaining renarration quality.
Multilingual Support: Creating efficient on-device language models that can handle cross-linguistic renarration, cultural context adaptation, and nuanced translation beyond simple word replacement.
AI Model Integration
- Vision-Language Models (VLMs) to interpret and renarrate multimedia content, making visual information accessible through descriptive renarration and context-aware explanations
- Language Models (LLMs) to generate user-specific renarrations, adjusting tone, complexity, depth, and cultural framing based on learned user preferences and stated needs
- Adaptive Learning Systems that improve renarration quality over time by understanding user feedback and interaction patterns while maintaining complete privacy
Project Deliverables
- Functional On-Device Prototype: A working system capable of performing automated renarration of diverse web content types, optimized for local device deployment
- Technical Architecture Documentation: Comprehensive report detailing model compression techniques, privacy-preserving design decisions, and performance optimization strategies
- User Experience Evaluation: Case studies and usability testing results illustrating how renarration improves accessibility, comprehension, and web utilization effectiveness across diverse user groups
Significance and Impact
Renarration is an essential function of human communication, as we instinctively adjust our explanations based on who we are addressing. This project seeks to replicate that adaptability through AI-driven solutions deployed with complete privacy protection. By implementing this assistance on-device, the project addresses growing concerns about digital privacy while tackling significant technical challenges in edge AI deployment.
The project bridges gaps in digital content consumption, making web information more context-aware, audience-specific, and universally accessible without compromising user privacy. Students will contribute to the crucial intersection of accessibility technology and privacy-preserving AI, developing skills in edge computing, model optimization, and human-centered design while creating solutions with immediate real-world applicability.
Offered by: Suzan Uskudarli & Onur Gungor
Project Description
Senior Project Description
The internet is a vast resource of information, yet not all web content is easily consumable by every user. This project focuses on developing an intelligent on-device web assistant that performs user-specific renarration of web content, automatically rearticulating information in ways that align with different user needs, contexts, and capabilities. Rather than simply providing content, this assistant serves as a sophisticated intermediary that helps users navigate, understand, and effectively utilize web resources. The aim is to make digital information more inclusive and adaptable, ensuring that content is not just available but truly understood by diverse audiences while preserving complete user privacy through local processing.
Renarration and Accessibility
Renarration is the process of transforming content to match the perspective, background, or situational needs of a given audience. This project positions the system as an intelligent assistant that understands both the user and the content, facilitating more effective web utilization. Accessibility in this context extends beyond traditional disabilities to include anyone who may struggle to engage with web content due to:
For example, a news article about a historical event could be renarrated differently for:
Technical Approach: On-Device Intelligence
The system will leverage state-of-the-art AI models deployed entirely on user devices, addressing critical technical and privacy challenges:
On-Device Processing Advantages
Privacy Preservation: All user interactions, browsing patterns, personal preferences, and renarration requests remain completely local. No sensitive data is transmitted to external servers, ensuring absolute privacy protection—particularly crucial when users access medical information, personal research, or sensitive topics.
Enhanced Accessibility: On-device processing eliminates internet dependency for core renarration functionality, providing consistent assistance even with poor connectivity. This democratizes access for users in regions with limited bandwidth, those on restrictive data plans, or in situations where connectivity is unreliable.
Reduced Latency: Local processing enables real-time content transformation without network delays, creating a seamless browsing experience where renarration happens instantly as users navigate web pages.
Core Technical Challenges
Model Optimization: Implementing efficient model compression and quantization techniques to fit capable Language Models and Vision-Language Models within mobile device memory and processing constraints while maintaining renarration quality.
Multilingual Support: Creating efficient on-device language models that can handle cross-linguistic renarration, cultural context adaptation, and nuanced translation beyond simple word replacement.
AI Model Integration
Project Deliverables
Significance and Impact
Renarration is an essential function of human communication, as we instinctively adjust our explanations based on who we are addressing. This project seeks to replicate that adaptability through AI-driven solutions deployed with complete privacy protection. By implementing this assistance on-device, the project addresses growing concerns about digital privacy while tackling significant technical challenges in edge AI deployment.
The project bridges gaps in digital content consumption, making web information more context-aware, audience-specific, and universally accessible without compromising user privacy. Students will contribute to the crucial intersection of accessibility technology and privacy-preserving AI, developing skills in edge computing, model optimization, and human-centered design while creating solutions with immediate real-world applicability.
Offered by: Suzan Uskudarli & Onur Gungor