← Back to Portfolio
Pinecone · demo

Bird Search: full-text meets visual search

A demo app combining Pinecone full-text search with multimodal vector search to find any of 2,000+ North American birds — by name or by what they look like.

Bird Search is a demo that pairs Pinecone Full-Text Search with multimodal vector search over a Wikipedia corpus of more than 2,000 North American bird species. You can search by keyword, describe what a bird looks like and find it by appearance, or combine both in a single query.

What it shows

  • A text tab with BM25 keyword search, phrase matching, and multi-field queries
  • A visual tab that finds birds by description, matching on appearance
  • A combined mode that applies hard text filtering with visual reranking in one API call
  • Lucene-style boolean operators, term boosts, and cross-field queries
  • Multimodal embeddings (Gemini Embedding 2) that put text and images in one vector space — no separate captioning pipeline

Built in Python with Streamlit and Pinecone.

Skills & Technologies

Pinecone Full-Text Search Multimodal Search Gemini Embedding Vector Search Semantic Search Streamlit Python
// stay in touch

Get Answering Machines in your inbox.

Occasional notes on AI, coding agents, and where this is all headed — plus the very occasional origami diagram.

Arjun Kirti Patel
Chicago, IL · folded with care
© 2026 Arjun Kirti Patel · MIT Licensed