UNplugged.ai: An Offline-First Multi-Modal Local LLM Workbench with Integrated Emotion Analytics

This paper has presented UNplugged.ai, an offline-first multi-modal AI workbench that consolidates six interaction modalities within a single Streamlit application backed by a locally-running Ollama inference daemon. The system demonstrates the feasibility of constructing privacy-preserving, locally-executable AI productivity tools using the Python open-source ecosystem without dependence on cloud inference infrastructure or vector database deployments. The direct context-stuffing approach to document grounding eliminates retrieval pipeline complexity at the cost of context window scalability, a tradeoff appropriate for moderate-length documents on hardware with sufficient memory bandwidth to serve the selected model efficiently. Limitations identified in the current implementation define a concrete roadmap for future enhancement. Immediate priorities include replacement of the Google speech recognition backend with a locally-executable alternative such as Whisper [18] or Vosk, correction of the PIL import error in the image tab, and implementation of a configurable chunking and retrieval pipeline supporting longer document contexts. Longer-term directions include structured multimodal image attachment support compatible with vision-capable Ollama models, persistent cross-session conversation storage, and a formal user study evaluating the system's usability and perceived trustworthiness relative to cloud-hosted alternative workflows.  

[1]  A. Narayanan and V. Shmatikov, "Robust De-anonymization of Large Datasets (How to Break Anonymity of the Netflix Prize Dataset)," in Proc. IEEE Symp. Security and Privacy, Oakland, CA, 2008, pp. 111–125. [2]  Ollama, "Ollama: Get up and running with large language models locally," 2023. [Online]. Available: https://ollama.com [3]  T. B. Brown et al., "Language Models are Few-Shot Learners," Advances in Neural Information Processing Systems (NeurIPS), vol. 33, pp. 1877–1901, 2020. [4]  A. Streamlit Inc., "Streamlit: The fastest way to build and share data apps," 2019. [Online]. Available: https://streamlit.io [5]  G. Gerganov, "llama.cpp: LLM inference in C/C++," GitHub repository, 2023. [Online]. Available: https://github.com/ggerganov/llama.cpp [6]  Y. Chang et al., "A Survey on Evaluation of Large Language Models," ACM Trans. Intelligent Systems and Technology, vol. 15, no. 3, pp. 1–45, 2024. [7]  R. Taori et al., "Alpaca: A Strong, Replicable Instruction-Following Model," Stanford CRFM Blog, 2023. [Online]. Available: https://crfm.stanford.edu/2023/03/13/alpaca.html [8]  A. Abid, A. Abdalla, A. Abid, D. Khan, A. Alfozan, and J. Zou, "Gradio: Hassle-free sharing and testing of ML models in the wild," arXiv preprint arXiv:1906.02569, 2019. [9]  P. Lewis, E. Perez, A. Piktus et al., "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks," Advances in Neural Information Processing Systems (NeurIPS), vol. 33, pp. 9459–9474, 2020. [10] H. Chase, "LangChain: Building applications with LLMs through composability," GitHub repository, 2022. [Online]. Available: https://github.com/langchain-ai/langchain