2026-09-22T00:11:08Zhttps://dspace.mit.edu/server/oai/requestoai:dspace.mit.edu:1721.1/1646712026-01-30T03:24:55Zcom_1721.1_7582com_1721.1_7581col_1721.1_131023
Han, Song
Huang, Siyong
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
2026-01-29T15:07:03Z
2026-01-29T15:07:03Z
2025-09
2025-09-15T14:56:24.389Z
https://hdl.handle.net/1721.1/164671
Large Language Models (LLMs) have proven highly effective for a wide range of natural language processing tasks, but their size and compute requirements often restrict their use to powerful cloud-based infrastructures. In recent years, significant progress has been made in shrinking LLMs while maintaining performance levels comparable to much larger models. We are approaching the point where the capabilities of massive, multi-billion parameter models can be realistically replicated on consumer-grade devices. This thesis builds upon that foundation by developing an AI-powered note-taking application that runs entirely offline, using only the compute resources available on a personal laptop. The application is designed to listen to lectures alongside the student and provide support in real-time—through transcription, notes generation, and enabling context-aware search. Achieving this level of interactivity locally introduces challenges in reducing end-to-end latency, which this project addresses through both model-level optimizations and the design of efficient prompting and inference algorithms. A demo of the app can be found on Youtube.
M.Eng.
Massachusetts Institute of Technology
In Copyright - Educational Use Permitted
Copyright retained by author(s)
https://rightsstatements.org/page/InC-EDU/1.0/
Cognify: An On-Device, AI-powered Learning Assistant
Thesis
application/pdf
Master
Master of Engineering in Electrical Engineering and Computer Science
Publication
unknown
unknown
<Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="849e7afd-6a92-4dfb-b283-518d7b54f655">
<Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type>
<Title>Cognify: An On-Device, AI-powered Learning Assistant</Title>
<PublishedIn>
<Publication>
</Publication>
</PublishedIn>
<PublicationDate>2025-09</PublicationDate>
<Authors>
<Author>
<DisplayName>Huang, Siyong</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
</Authors>
<Editors>
</Editors>
<Publishers>
<Publisher>
<DisplayName>Massachusetts Institute of Technology</DisplayName>
<OrgUnit />
</Publisher>
</Publishers>
<License>https://rightsstatements.org/page/InC-EDU/1.0/</License>
<Abstract>Large Language Models (LLMs) have proven highly effective for a wide range of natural language processing tasks, but their size and compute requirements often restrict their use to powerful cloud-based infrastructures. In recent years, significant progress has been made in shrinking LLMs while maintaining performance levels comparable to much larger models. We are approaching the point where the capabilities of massive, multi-billion parameter models can be realistically replicated on consumer-grade devices. This thesis builds upon that foundation by developing an AI-powered note-taking application that runs entirely offline, using only the compute resources available on a personal laptop. The application is designed to listen to lectures alongside the student and provide support in real-time—through transcription, notes generation, and enabling context-aware search. Achieving this level of interactivity locally introduces challenges in reducing end-to-end latency, which this project addresses through both model-level optimizations and the design of efficient prompting and inference algorithms. A demo of the app can be found on Youtube.</Abstract>
<Access xmlns="http://purl.org/coar/access_right"
>
</Access>
</Publication>