Semantic Search Engine

A small app that searches a list of startups by meaning.

• Neural search reads each startup's description and finds similar ones.
• Keyword search matches your exact term in the description.

Run locally

Local runs use a throwaway Qdrant in Docker — no cloud account needed.

Prerequisites: Python 3.11, Docker

# 1. Environment
python -m venv .venv
source .venv/bin/activate

# 2. Dependencies
pip install poetry
poetry install

# 3. Dataset
wget https://storage.googleapis.com/generall-shared-data/startups_demo.json -P data/

# 4. Start Qdrant + the service
docker-compose -f docker-compose-local.yaml up

# 5. Load the data
python -m qdrant_demo.init_collection_startups

Then open http://localhost:8000/.

Larger dataset (Crunchbase)

To index a bigger set of companies, get a Crunchbase API key, then:

wget 'https://api.crunchbase.com/odm/v4/odm.tar.gz?user_key=<CRUNCHBASE-API-KEY>' -O odm.tar.gz
tar -xvf odm.tar.gz
mv odm/organizations.csv ./data
python -m qdrant_demo.init_collection_crunchbase

What's inside

Software stack
Qdrant	Vector database and search engine with full-text and semantic capabilities.
mxbai-embed-large-v1	The embedding model that turns startup data into vectors.
Qdrant/bm25	The sparse model behind keyword search.
Qdrant Cloud inference	Embeds the query server-side, so the app ships no local model.
React (Vite)	The frontend, styled with the Qdrant design system.

Component
init_collection_startups.py	Loads startup data into a Qdrant collection (with a text index for keyword search).
init_collection_crunchbase.py	Same, for the larger Crunchbase dataset.
neural_searcher.py	Semantic search: embeds the query and returns the nearest startups.
text_searcher.py	Keyword search: full-text match on the description field.
service.py	FastAPI app exposing GET /api/search?q&neural, also serving the built frontend.
config.py	Reads env vars (Qdrant URL/key, collection, embeddings model). Text field defaults to description, overridable via TEXT_FIELD_NAME.

Deploy (Qdrant Cloud)

A deployed instance searches a Qdrant Cloud collection instead of a local one. Load the collection first (the init_collection_* scripts above, pointed at your cluster), then set these environment variables wherever you deploy:

Variable	Value
QDRANT_URL	your Qdrant Cloud endpoint (https://…:6333)
QDRANT_API_KEY	your Qdrant Cloud API key
COLLECTION_NAME	the collection to search (e.g. startups)

Option A — one container (Railway or any Docker host)

The Dockerfile builds the React frontend and runs FastAPI serving it, so the whole demo is a single service. On Railway: New → Deploy from GitHub repo, pick this repo, add the variables above. The container binds $PORT automatically.

Option B — frontend on Vercel, API on the container

To embed the demo behind a static link, host the UI on Vercel and keep the API on the container from Option A:

1. Import this repo on Vercel with Root Directory = frontend (Vite is auto-detected).
2. Add one env var: VITE_API_BASE = the container's URL. The frontend then calls that API cross-origin; it defaults to same-origin, so Option A is unaffected.