retrieval-augmented-generation-rag

Learning notes on fundamentals of retrieval augmented generation.

Large Language Model (LLM)

LLM Generation Challenges

• Limited with data they are trained on
• Hard to adapt
• Answer does not have source
• Answer is out of date and changes over time
• Data is specific to business

Vector Database

• Mathematical representation of strucutured and unstructured data similar to an array
• Query vector database, we get back embeddings which includes relevant data, merged into the original prompt and fed into LLM
• As new data comes in, vector embeddings gets updated

Vector

An array of data that gets stored into database in numerical form.

Embeddings

A multi demensional representation of vector data.

AI Agents

Enables compound AI system by combining LLM models and independent modules to answer a user query.

Worries and concerns

• Taking orders, processing refunds
• Hallucinations
• Inaccurate results
• Perpetuate bias

References

https://www.youtube.com/watch?v=T-D1OfcDW1M&list=PLn_EYvIwertwJEf3h4iweYN7iWEkf6x2s