title	photon-route
emoji	💡
colorFrom	indigo
colorTo	blue
sdk	docker
app_port	7860
pinned	false
license	mit
short_description	Continuous-variable photonic retrieval (Strawberry Fields).
sleep_time	3600

photon-route

A continuous-variable (CV) photonic retrieval research artifact.

Each document and query is encoded as a small Strawberry Fields program — words contribute squeezing and displacement parameters to alternating bosonic modes, sentence length controls a beam-splitter mixing layer, and the resulting Gaussian state is the document's representation in Fock space. Retrieval ranks documents by closed-form Gaussian-state fidelity (Banchi-Braunstein-Pirandola) between query and document states.

This is the CV sister to qrouter, which takes the discrete-variable (DV / qubit-gate-model) path. Both share the same fixture and ranking interface; the substrate is what changes.

Status

Day-1 scaffold. The pipeline is end-to-end correct in shape; numerical results are placeholders until:

• The encoding is replaced with a learned parameter map (today's params are SHA-256-derived from word identity)
• A real eval set (~200–500 query-doc pairs over arXiv quant-ph) is built
• A non-Gaussian element (PNR measurement on a subset of modes, conditioning the rest) is added to escape Gaussian classical-simulability
• The same program is run on Borealis via Xanadu Cloud for the simulated-vs-measured comparison

See docs/design.md for the research diary.

Quickstart (local)

uv venv && source .venv/bin/activate
uv pip install -e ".[photonic,dev]"
photon-route "what happens when light hits a barrier"

Or run the HTTP server:

uvicorn photon_route.http_server:app --port 7860
curl 'http://127.0.0.1:7860/rank?q=quantum+entanglement&top_k=3'

Endpoints (HTTP)

• GET / — JSON banner with backend mode
• GET /health — {"ok": true, "backend": "..."} where backend is gaussian if Strawberry Fields imports cleanly, else stub
• GET /rank?q=...&top_k=N — ranked fixture documents
• GET /version, GET /docs — metadata + FastAPI Swagger

License

MIT.