A Hybrid Framework Integrating LLM and RAG for Personalized and Adaptive Skill Roadmap Generation in Career and Learning Pathways

As technology continues to develop rapidly due to the availability of information technology, learners are in need of developing their skills. Hence, learners frequently encounter unstructured material and finding simple routes through which they can learn new skills. This research provides a framework that integrates AI-enabled tools along with large language models (LLMs) into a retrieval-augmented generation (RAG) framework, for enabling the generation of personalized skill development maps to provide required information and materials. By using these methods, users can pursue learning pathways across multiple domains, based on each user’s current level of skill, their individual goal, with their available time. With the locally hosted Mistral-7B-Instruct model with the Ollama framework, this study prioritizes privacy, efficiency, and the ability for users to create their learning pathways offline. Contextual retrieval capabilities are provided through a Pinecone vector database that uses 384-dimensional sentence transformer embeddings built with all-MiniLM-L6-v2, which supports a multi-phase pipeline that includes dataset curation, prompt preparation, retrieval, and the generation of an interactive roadmap. Through this research, we contribute to a privacy preserving solution with salability. It will helpful for adaptive career development and also the institutional training programs. It reduces the time of learning track discovery to 65% compared to manual approach. The personalization quality and experimental evaluation accuracy is 80%, precision is 82%and measured recall is 79%. The overall average user satisfaction score has increased to 4.2 out of 5.0.