RAG based AI representative for creators
The potential of AI to enhance creative applications is huge, yet it often remains largely accessible to tech giants and startups with clear project-market fits. For individual creators and artists, the high computing costs and technical barriers of hosting and developing large language model (LLM)-based works present significant challenges. This project explores how LLMs can be leveraged on a smaller, more personal scale, transforming them into powerful creative tools tailored to individual needs.
As part of this exploration, I developed an LLM-based portfolio chatbot using a Retrieval-Augmented Generation (RAG) approach. The chatbot retrieves personal information and portfolio content from a markdown-based rich media database and runs on a compact 8B LLM model locally, hosted on a Google Colab node. By employing a tailored set of prompt templates, the chatbot serves as a virtual representative capable of answering inquiries about my work in a visually rich and stylistically engaging manner. Key features include:
Cost-Effective Local Deployment: Hosting smaller LLM models with personal data on personal-level servers ensures accessibility for individual creators.
Rich Media Integration: The chatbot can retrieve and display images from the database, showcasing work visually alongside textual responses.
Personalized, Stylistic Responses: Using LangChain’s LLM Chain, the chatbot provides accurate and engaging answers through role-playing prompt templates.
By enabling creators to access AI on an individual scale, this chatbot attempt to challenge the dominant narrative of AI as solely a large-scale industrial tool, demonstrating its potential to foster personalized and meaningful interactions.
