Skip to content

Semantic "Vibe" Search

Find places by their qualitative "vibe" rather than keywords or coordinates.

Traditional search finds "coffee" or "gym". Vibe search uses vector embeddings to understand the semantic meaning of places.

It finds places that feel like your query, even if they don't contain the exact words.

  • "Quiet place to read" -> matches Parks, Libraries, Cozy Cafes
  • "Late night bonding" -> matches Pubs, 24/7 Diners, Rooftops

How It Works

  1. Embed: Convert text (place names, hints, descriptions) into a vector.
  2. Index: Store place vectors.
  3. Query: Convert user query into a vector.
  4. Compare: Find nearest neighbors using cosine similarity.

Step 1: Build the Index

You need sentence-transformers installed.

from chora.core import PlatialGraph
from chora.search import build_place_index, PlaceIndex

# Load your populated graph
graph = PlatialGraph.load("my_city.json")

# Build index (computes embeddings for all places)
# Uses 'all-MiniLM-L6-v2' by default (small & fast)
index = build_place_index(graph)

# Save for later
index.save("place_index.json")

Step 2: Search by Vibe

from chora.search import vibe_search

# Load index if needed
# index = PlaceIndex.load("place_index.json")

# 1. Productive vibe
results = vibe_search(
    graph, 
    "place to get deep work done with coffee",
    index=index
)

for extent, score in results[:3]:
    print(f"{score:.2f} | {extent.name} ({extent.extent_type})")
# 0.82 | Third Space Library (public)
# 0.79 | Ozone Roasters (cafe)
# 0.65 | WeWork Lobby (coworking)

# 2. Romantic vibe
results = vibe_search(
    graph,
    "romantic evening spot with a view",
    index=index
)
# 0.85 | Sky Garden (viewpoint)
# 0.78 | Southbank Walk (path)

Step 3: Find Similar Places (Recommendation)

"If you like The Book Club, you might like..."

from chora.search import find_similar_places

# Get reference place
book_club = graph.get_node_by_name("The Book Club")

# Find lookalikes
recommendations = find_similar_places(
    graph,
    reference_extent=book_club,
    top_k=3
)

Advanced: Custom Embeddings

You can enrich the embeddings by including more context:

from chora.search import embed_extent

# Custom embedding that prioritizes "activities"
embedding = embed_extent(
    place, 
    graph, 
    include_encounters=True,  # Include "visited 50 times"
    model_name="all-mpnet-base-v2"  # Larger, more accurate model
)

CLI Usage

Currently available as a library function. CLI integration coming in v0.9.5.

# Quick script
from chora.cli import get_graph
from chora.search import vibe_search

graph = get_graph()
print(vibe_search(graph, "cozy spot"))

Next Steps