feat: add semantic search marimo notebook

[jhamon]

Summary

Adds a new marimo notebook demonstrating semantic search with Pinecone, converted and significantly expanded from the existing docs/semantic-search.ipynb. The notebook uses Pinecone's Integrated Inference with the multilingual-e5-large model to demonstrate cross-lingual semantic search across English and Spanish sentences.

Changes

• New notebook docs/semantic-search.py (marimo format) with:
  • Pinecone SDK 9.0.1 API (pc.indexes.*, pc.index(), updated search signature)
  • multilingual-e5-large embedding model for cross-lingual retrieval
  • Refactored dataset prep: filter_pairs + extract_sentences(lang) to produce both English and Spanish records from Tatoeba
  • to_records parameterized on column name with ID prefixes for multi-language upsert
  • mo.ui.table for dataset inspection, mo.status.progress_bar replacing tqdm, mo.ui.run_button for safe index deletion
  • Interactive query section with mo.ui.text and mo.ui.radio for language filter
  • Language filtering section demonstrating metadata filters scoped to en/es
  • Prose interspersed between code cells narrating the process
  • "Meaning Over Keywords" and "How It Works" sections explaining model selection and cross-lingual retrieval
• pyproject.toml: pins notebook dependencies (datasets==3.5.1, pinecone==9.0.1, numpy, tqdm)

Test Plan

• Notebook runs end-to-end with a valid PINECONE_API_KEY
• Index creation, upsert, and query cells execute without errors
• Cross-lingual queries return results in both languages
• Language filter correctly scopes results to en or es
• Interactive query input updates results on change
• Delete button safely removes the index

🤖 Generated with Claude Code

---

Note

Low Risk
Low risk: adds a new documentation notebook only, with no changes to production code paths; the main impact is on users running the example (it creates/deletes a Pinecone index).

Overview
Adds a new docs/semantic-search.py Marimo notebook that walks through building a semantic search demo with Pinecone Integrated Inference, including index creation for multilingual-e5-large, dataset filtering/record preparation for English+Spanish, batched upsert_records, and index.search with optional lang metadata filtering.

The notebook also adds interactive UI elements for querying and a run-button gated cleanup step to delete the created index.

^{Reviewed by Cursor Bugbot for commit a233d4b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Incorrect index.search call shape

High Severity

index.search passes top_k, inputs, and filter as top-level keyword arguments without a query dict. For integrated text search, Pinecone v9 still expects query={"inputs": {"text": ...}, "top_k": ..., "filter": ...}, so these calls will raise a TypeError or fail validation and break all query cells.

 

^{Reviewed by Cursor Bugbot for commit 72b422c. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

There are 2 total unresolved issues (including 1 from previous review).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fc1dc00. Configure here.}