What you'll learn

Learn to recognize when queries are producing poor results.
Learn to use a large language model (LLM) to improve your queries.
Learn to fine-tune your embeddings with user feedback.

Course recap

PRO

This course, built in partnership with Chroma, teaches advanced retrieval techniques that go beyond simple semantic similarity search in RAG applications. It addresses common failure modes where naive vector search returns distracting or irrelevant results, and provides practical methods to improve retrieval quality.

Concept map

Key concepts

Retrieval-Augmented GenerationThe overall pattern of retrieving relevant documents to provide context to an LLM for answering queries
EmbeddingsDense vector representations of text using sentence transformer models; the course emphasizes understanding embedding model context windows (256 tokens for sentence transformers)
Query expansionUsing LLMs to augment queries for better retrieval, either by generating hypothetical answers or multiple related queries
Cross-encoder re-rankingScoring retrieved results for relevancy using a model that takes query-document pairs as input
Embedding adaptersLightweight trainable matrices that transform query embeddings based on user feedback to improve retrieval
DistractorsRetrieved results that are not relevant to the query but can mislead the LLM during synthesis

Lesson highlights

1.Introduction — Overview of advanced retrieval techniques and course structure
2.Overview of embeddings-based retrieval — End-to-end RAG system with Chroma, chunking strategies, embedding functions
3.Pitfalls of retrieval — Visualization with UMAP, failure modes, irrelevant results, geometric intuition
4.Query expansion — Generated answers (HyDE-style), multiple query generation, embedding space visualization
5.Cross-encoder re-ranking — Bi-encoder vs cross-encoder, scoring relevancy, mining the long tail of results
6.Embedding adapters — Training lightweight adapters using user feedback, cosine similarity loss, dimension importance
7.Other techniques — Fine-tuning embedding models, fine-tuning LLMs for RAG, intelligent chunking

About this course

Information Retrieval (IR) and Retrieval Augmented Generation (RAG) are only effective if the information retrieved from a database as a result of a query is relevant to the query and its application.

Too often, queries return semantically similar results but don’t answer the question posed. They may also return irrelevant material which can distract the LLM from the correct results.

This course teaches advanced retrieval techniques to improve the relevancy of retrieved results.

The techniques covered include:

Query Expansion: Expanding user queries improves information retrieval by including related concepts and keywords. Utilizing an LLM makes this traditional technique even more effective. Another form of expansion has the LLM suggest a possible answer to the query which is then included in the query.
Cross-encoder reranking: Reranking retrieval results to select the results most relevant to your query improves your results.
Training and utilizing Embedding Adapters: Adding an adapter layer to reshape embeddings can improve retrieval by emphasizing elements relevant to your application.

Who should join?

Anyone who has intermediate Python knowledge and wants to learn advanced retrieval techniques for retrieving data from their vector database.

Advanced Retrieval for AI with Chroma

What you'll learn

Course recap

Concept map

Key concepts

Lesson highlights

About this course

Who should join?

Course Outline

Elevate your learning experience with Pro

Instructor

Anton Troynikov

Advanced Retrieval for AI with Chroma

Want to learn more about Generative AI?