ACUNAO AI Chatbot
An AI assistant as a domain expert of a project's documents and has the ability to answer users’ queries accurately. It assists users with organizing their projects and acts as a 'second brain'.
- python
- ai
- ds
- ui/ux
Last modified:
→ Click here to view the project repository
Problem Statement
Research labs may handle large volumes of documents that may comprise of complex literatures and confidential documents. The manual process of examining and managing these literatures and documents can be time-consuming and prone to errors. In addition, parsing through literatures requires expertise in the field, which may not be the case when labs of differing expertise are collaborating.
Proposed Solution
- Develop an Artificial Intelligence (AI) system for document processing to extract information from documents effectively
- Use the Retrieval Augmented Generation (RAG) framework to enhance the quality of the AI assistant’s responses
- Build trust between the users and AI assistant by providing references to the responses and admitting to lack of knowledge if the AI assistant does not know the answer to a question
- Ensure adaptability across different research environments and integration into lab workflows
Relevance
Due to the rapid growth of AI, there has been a drastic increase in AI-driven solutions. These tools have mostly been developed to appeal to the general public. The proposed AI assistant solution offers a promising approach for research labs to manage and retrieve information from documents, which as a result, becomes a bridge for non-experts to understand literatures outside their field. This helps with collaboration between labs and enhances efficiency in research labs by automating information management and improving data accessibility.
Output Generation
The response from the model contains two parts: the documents retrieved with metadata, and the overall summary of the answer. The fastest response time is 8.98 seconds on a Macbook Air M3 16GB Metal GPU-enabled laptop. The average response time is <20 seconds and it increases based on the context size and complexity of response.
GUI
The GUI for the prototype is built using the Streamlit Python framework. It is the simplest solution for an interface at the current stage of the project.
