Teaching Assistant - Powered by LLM and RAG

• 

  Project Page

Inspiration

Teaching Assistant's inspiration is rooted in addressing educational disparities exacerbated by COVID-19. While students in larger institutions and urban areas enjoyed advanced online learning, those in smaller or economically challenged setups lacked such resources. This digital gap fueled our commitment to building a locally and cloud-operable system, accessible to all. Our open-source model aims not only to empower educational institutions but also to directly benefit students. By making quality learning support available to everyone, our Teaching Assistant levels the educational playing field, ensuring inclusivity for both institutions and individual learners.

What it does

Teaching Assistant, fueled by LLM (Language-Model) and RAG (Retrieval-Augmentation-Generation), assumes the role of an instant personal assistant for students. The system diligently processes uploaded books, transforming them into vector data using Google Generative AI Embeddings. Efficient storage and retrieval of this vector data are facilitated through the FAISS library, elevating the overall performance of the assistant. When a user submits a query, the system seamlessly converts it into a vector query, drawing upon the local database to generate human-readable answers through the LLM model. The architecture excels in delivering personalized assistance across multiple languages and subjects, positioning itself as a potent tool for students seeking educational support.

How we built it

The creation of the Teaching Assistant project was an artful blend of cutting-edge technologies. Google Generative AI Embeddings took the lead in text processing, while FAISS played a pivotal role in optimizing the storage and retrieval of vector data. The Gemini-Pro model lent its prowess to profound analysis and comprehension of book content. Aiding in efficient text segmentation, the recursive character text splitter played a crucial role. The resulting architecture, known as RAG + LLM, seamlessly integrates these components, ensuring the development of a robust and efficient learning assistant.

Challenges we ran into

The journey of constructing Teaching Assistant presented its unique set of challenges. Integrating diverse advanced technologies and ensuring their cohesive operation presented initial hurdles. The process of semantic searching, converting to best vector DB models for multilingual support and optimizing overall performance demanded dedicated efforts. Overcoming these challenges necessitated collaborative problem-solving, adaptability, and a continuous refinement of our approach.

Accomplishments that we're proud of

Our team derives immense pride from the successful development of a learning assistant that directly addresses the real-world challenges faced by students. The achievement of delivering instant assistance in multiple languages, breaking down language barriers, and providing a versatile tool for educational support stands as a testament to our unwavering commitment to excellence. The adept utilization of advanced technologies, including Google Generative AI Embeddings, FAISS, and the Gemini-Pro model, highlights our dedication to creating a robust and innovative solution.

What we learned

The odyssey of creating Teaching Assistant imparted invaluable lessons. From navigating the intricacies of integrating generative AI embeddings to optimizing vector data storage with FAISS, each step significantly contributed to our understanding of cutting-edge technologies. Collaborative problem-solving and the ability to adapt our approach in response to challenges enhanced our team's skills and knowledge.

What's next for Teaching Assistant - Powered by LLM and RAG

The future trajectory of Teaching Assistant involves continuous refinement and expansion. We aspire to augment language support, incorporating additional languages to make educational assistance even more accessible. A key focus involves fine-tuning the model for specific purposes, enhancing the quality of assistance. Additionally, we aim to refine the user interface, ensuring a seamless and intuitive experience. Ongoing updates will concentrate on improving the efficiency of the learning assistant, incorporating valuable user feedback, and exploring opportunities for further innovation in educational technology. Teaching Assistant is not merely a project; it embodies our unwavering commitment to the continuous enhancement of educational support for students worldwide.

Built With

• embeddings
• faiss
• gemini
• langchain
• steamlit