Updated LlamaIndex chatbot sample

[Allen Kim (allenkim0129)]

Description

Updated LlamaIndex-CosmosDBNoSql-RAG-Chatbot-Sample.ipynb to align with the new search capabilities introduced in run-llama/llama_index#20871, which adds 7 search types and a fully parameterised query engine to llama-index-vector-stores-azurecosmosnosql.

Changes

Notebook format

• Converted from a Jupyter #%%-style Python script to a proper .ipynb file with separate, individually-runnable code and markdown cells.

Credentials

• Replaced bare os.getenv(...) calls with explicit "YOUR_AZURE_..." placeholder fallbacks so the notebook is self-documenting and works out of the box without a .env file.

Dependencies (Section 1)

• Added explicit %pip install cells for llama-index-vector-stores-azurecosmosnosql, gradio, and python-dotenv.

Full Text Search store (Section 6)

• Added a dedicated fts_store container configured with full_text_policy and fullTextIndexes in the indexing policy, required for full text and hybrid search types.

Search types showcase (Section 7)

Demonstrated all 7 search types via direct store.query() / fts_store.query() calls using AzureCosmosDBNoSqlVectorSearchType:

Section	Search Type	Key detail
7a	VECTOR	Default nearest-neighbour, ordered by VectorDistance
7b	VECTOR_SCORE_THRESHOLD	Adds threshold kwarg to filter low-similarity results
7c	FULL_TEXT_SEARCH	FullTextContains WHERE — no query_embedding passed
7d	FULL_TEXT_RANKING	ORDER BY RANK FullTextScore(...) — single-field and multi-field (RRF) examples; no query_embedding passed
7e	HYBRID	ORDER BY RANK RRF(FullTextScore, VectorDistance)
7f	HYBRID_SCORE_THRESHOLD	Hybrid RRF + client-side threshold filter
7g	WEIGHTED_HYBRID_SEARCH	Hybrid RRF + weights list for per-component client-side re-ranking

Advanced query options showcase (Section 8)

Added individual cells demonstrating each new query kwarg:

Section	Kwarg	Description
8a	where	Raw CosmosDB SQL WHERE predicate for pre-filtering
8b	offset_limit	OFFSET x LIMIT y pagination (replaces TOP for hybrid types)
8c	projection_mapping	Map Cosmos DB document fields to LlamaIndex node metadata keys
8d	return_with_vectors	Include raw embedding vectors in returned nodes

Gradio chatbot (Section 9)

• Moved to the final section; wired to the default index.as_query_engine() (vector search).
• Improved label copy and elapsed-time formatting.

---

Type of Change

• Documentation / sample update (non-breaking)

Related

• Upstream LlamaIndex PR: run-llama/llama_index#20871

Testing

Run the notebook top-to-bottom against a live Azure Cosmos DB for NoSQL account with:

• Vector search preview feature enabled
• Full text search preview feature enabled (required for Sections 6, 7c–7g, 8b)