Automatic Speech Recognition (ASR) models in the context of transcription of endangered languages have been proven to significantly cut down word error rates (WER) when it comes to documentation of the languages. Although research has been done on different types of partitioning of the data when it comes to training the model and then subsequently testing the model, the "genre" (reading of the Bible, children's stories, Wikipedia, etc.) of the data has never been evaluated when it comes to WER. Thus, we aim to train multiple models independently on single types of data in order to investigate if any model performs better on any singular corpora type. We do this by examining the already existing FormosanBank, a large-scale data-driven project dedicated to the preservation and revitalization of the Indigenous Formosan languages of Taiwan. These languages, which form a significant part of the Austronesian language family, are endangered, with some facing the risk of extinction. ASR models have been generally trained on FormosanBank as a whole (baseline), but this study aims to break up the dataset by type to train in order to compare it to the baseline. 

Modern home networks consist of many user and Internet-of-Things devices, including smartphones, cameras, televisions, and many other types. Existing tools for analyzing this traffic, such as Wireshark, require a high level of technical knowledge, leaving non-technical users unable to tell what their devices are doing or why their network is slow. This study proposes a LoRA fine-tuned Qwen3-8B model that uses a capture of a device's traffic and explains what the device is doing in simple terms for non-technical users. From a given traffic capture, a data pipeline is used that converts network flow data into plain language summaries, and combines the plain-language summaries with a hand-written non-technical explanation, which is then used to fine-tune the model. The expected result is for the fine-tuned model to produce a clear, accurate, non-technical description of what is occurring in a home network.