GemAPI RAG Based search engine

Inspiration

Large Language Models (LLMs) like Gemini 1.5 Pro are incredible, but they often miss out on specific contextual information crucial for businesses and technical fields. To address this, my project integrates LLMs with real-time data from external sources such as PDFs and text files. This integration enhances the model's ability to leverage the most up-to-date information, ensuring no context is left behind.

What it does

A project that I've been working on a Retrieval-Augmented Generation (RAG) System tailored for the latest advancements in language understanding, demonstrated on the "Leave No Context Behind" paper published by Google on April 10, 2024.

How we built it

Utilizing the LangChain framework, this RAG system retrieves external documents—in this case, the profound insights of the "Leave No Context Behind" paper—and dynamically incorporates this data into Gemini 1.5 Pro’s response generation process. This approach significantly enriches the answers provided by the AI, making it a powerful tool for research and information retrieval. We can use the embeddings created by LLMs from sentence-transformers to the given pdf input and we have to store those embeddings to the ChormaDB database for further matching of query embeddings with these embeddings.

Challenges we ran into

RAG based approach when we have infinite amount of data because we need to create embeddings for the total data which needs more computational power and storage.

Accomplishments that we're proud of

What we learned

we learned that we can also use the LLMs for our purpose and we can update those models in our interested areas to get valuable insights from our data.

What's next for GemAPI RAG Based search engine

Built With

• chromadb
• embeddings
• gemini-api
• lang-chain
• llms
• sentence-transformers