Cleanup Assistent

Introduction The Cleanup-Assistant project is designed to modernize the way people organize and locate their belongings. This innovative project leverages advanced language models like gpt-oss to create a user-friendly and highly efficient virtual assistant that can answer questions such as "Where did I put this?" or "Which piece matches with which other one?" However foremost, beyond providing a practical and usable smart application, the project aims to revolutionize the contextual understanding of large language models (LLMs), in particular gpt-oss. Features and Benefits The Cleanup-Assistant offers a range of features that make it a standout project:

1. Natural Language Interaction: Users can interact with the assistant using everyday language, making it accessible and easy to use.
2. Efficient Token Usage: The project has successfully reduced by half the required number of tokens to be processed by gpt-oss using the treatment_data_tool, making the system more efficient.
3. Scalability and Modifiability: The architecture is designed to be easily modifiable and expandable, e.g. with additional capabilities, like applying it to digital items like emails or files on a computer. Evaluation Criteria
4. Application of gpt-oss: The project effectively applies the advanced capabilities of gpt-oss, showcasing its strengths in natural language understanding, general world knowledge, and robustness against variations in terminology. This unique application highlights the model's potential in a practical and impactful way.
5. Design: The user experience is thoughtfully designed, with a simple browser-based interface that includes branding elements and can be easily extended with standard web technologies. The architecture, based on agents and light weight state-of-the-art AI-backend technologies, ensures the system is both modifiable and expandable. Agentic capabilities implemented include:
6. save_data_tool: Takes an array of strings of context information and saves it to a vector
7. space.search_objects_tool: Queries a vector database to find possible known locations for a given query. It returns the closest matches based on the query string.
8. treatment_data_tool: Takes an array of strings of context information and compresses it (to make better use of the availbale context size)
9. Potential Impact: The project's core innovation lies in its technical foundations, which drastically reduce the resource requirements (energy) for using LLMs. This not only makes the service more affordable but also contributes to sustainability and climate protection. The techniques developed can be applied to any AI application involving LLMs. This makes the societal and technical impact significant, as LLMS are used ubitiquously.
10. Novelty of the Idea: The Cleanup-Assistant introduces new techniques for context optimization in LLMs, enriching existing approaches with a fresh perspective. The concept of the Cleanup-Assistant emerged from our brainstorming sessions and has not been encountered in this form before. Conclusion The Cleanup-Assistant project is a unique and innovative solution that leverages the strengths of advanced language models to provide a practical and impactful application. Its thoughtful design, significant potential impact, and novel approach make it a strong contender.

Built With

• langchain
• ollama
• qdrant