Rank products 19% better than keyword search

OpenSearch doc-v3-gte adds learned search terms to each product so shoppers can find it using different words. SIE is the open-source inference engine that serves the model; keep your existing search index.

Get started Run in your cloud

OpenSearch doc-v3-gte on SIE

Top 3 are exact matches

ryobi flashlight

1 Ryobi RFL905 160-Lumen … Cree LED Flashlight … Exact
2 Ryobi One+ P704 18V … Cordless Flashlight w/ Rotating Head … Exact
3 RYOBI P705 One+ 18V … LED 130 Lumen Flashlight … Exact

BM25, same search Replacement bulbs at ranks 1–3

Keyword search ranks the accessories above the product

BM25, Elasticsearch English analyzer

No scooter in the top 3

pride mobility scooter

1 Pride Mobility Cup Holder Folding Go-Go Scooter C-Style Molded Seat Armrest Complement
2 Pride Mobility Scooter Rear Basket Center Support with Free Holding Clevis Pin Complement
3 Clip Cane Holder for Pride Mobility Scooters (Set of 2) Complement

The first scooter labelled Exact is at rank 11. Shoppers must scroll past accessories to find the product.

BM25, Elasticsearch English analyzer

Bulbs instead of flashlights

ryobi flashlight

1 Ryobi Ridgid OEM 780287001 18V Flashlight Work Light Bulb (2 Pack) Complement
2 Podoy 780287001 Flashlight Bulbs for … Ryobi Ridgid 18V Flashlight … Complement
3 Ryobi 780287001 Light Bulbs for 18V Flashlight 4PK Complement

All three results are replacement bulbs. A shopper looking for a flashlight still has no product to buy.

Baymard Institute 56% of ecommerce sites fail to support shoppers’ search needs
Poor-performing ecommerce Search leads to frustrating search experiences, time wasted refining queries, and abandonments…
29 April 2026

19% better product ranking than keyword search, on 8,956 Amazon searches

Run this eval

Product ranking quality

BM25 = 100 ↑

SIE OpenSearch doc-v3-gte

SIE OpenSearch doc-v3-gte 119

Keyword search (BM25)

Keyword search (BM25) 100

Model API cost to index

USD, 482,105 products ↓

Keyword search (BM25)

$0 Keyword search (BM25)

SIE OpenSearch doc-v3-gte

SIE OpenSearch doc-v3-gte $3.38

Evaluation

Tested: 8,956 Amazon shopping searches across 482,105 products
Scored: nDCG@10, using human relevance labels
Prices: SIE Serverless and provider API list prices

Send the search and your listings, rank by shared terms

from sie_sdk import SIEClient
client = SIEClient(api_key="sk-sie-…")
query = "querywayfair comforters"
documents = [
    "document 1···wayfair basics® comforter set···",
    "document 2···wayfair sleep medium weight all season polyester down alternative comforter···",
    "document 3···wayfair basics® reversible quilt set···",
    "document 4wayfair basics single curtain rod. Curtain Hardware & Accessories",
    "document 5wayfair basics peva shower curtain liner",
    "document 6wayfair basics round single curtain rod. Curtain Hardware & Accessories",
]
model = "modelopensearch-project/opensearch-neural-sparse-encoding-doc-v3-gte"
q = client.encode(model, [{"text": query}], output_types=["sparse"], is_query=True)[0]["sparse"]
docs = client.encode(model, [{"text": d} for d in documents], output_types=["sparse"], is_query=False)
print(q)
print(docs)

import { SIEClient } from '@superlinked/sie-sdk';

const client = new SIEClient(undefined, { apiKey: 'sk-sie-…' });

const query = "wayfair comforters";
const documents = [
  "···wayfair basics® comforter set···",
  "···wayfair sleep medium weight all season polyester down alternative comforter···",
  "···wayfair basics® reversible quilt set···",
  "wayfair basics single curtain rod. Curtain Hardware & Accessories",
  "wayfair basics peva shower curtain liner",
  "wayfair basics round single curtain rod. Curtain Hardware & Accessories",
];

const model = 'opensearch-project/opensearch-neural-sparse-encoding-doc-v3-gte';
const [q] = await client.encode(model, [{ text: query }], {
  outputTypes: ['sparse'],
  isQuery: true,
});
const docs = await client.encode(
  model,
  documents.map((text) => ({ text })),
  { outputTypes: ['sparse'], isQuery: false },
);
console.log({ query: q, documents: docs });

# Encode the query with is_query, then the listings; save native sparse vectors.
curl https://api.superlinked.com/v1/encode/opensearch-project%2Fopensearch-neural-sparse-encoding-doc-v3-gte \
  -H "Authorization: Bearer sk-sie-…" \
  -H "Content-Type: application/json" \
  -d "{\"items\":[{\"text\":\"wayfair comforters\"}],\"params\":{\"output_types\":[\"sparse\"],\"options\":{\"is_query\":true}}}" \
  -H "Accept: application/json" -o query.json

curl https://api.superlinked.com/v1/encode/opensearch-project%2Fopensearch-neural-sparse-encoding-doc-v3-gte \
  -H "Authorization: Bearer sk-sie-…" \
  -H "Content-Type: application/json" \
  -d "{\"items\":[{\"text\":\"···wayfair basics® comforter set···\"},{\"text\":\"···wayfair sleep medium weight all season polyester down alternative comforter···\"},{\"text\":\"···wayfair basics® reversible quilt set···\"},{\"text\":\"wayfair basics single curtain rod. Curtain Hardware & Accessories\"},{\"text\":\"wayfair basics peva shower curtain liner\"},{\"text\":\"wayfair basics round single curtain rod. Curtain Hardware & Accessories\"}],\"params\":{\"output_types\":[\"sparse\"],\"options\":{\"is_query\":false}}}" \
  -H "Accept: application/json" -o listings.json

Build the "Sparse embeddings" capability into my app using the Superlinked Inference Engine (SIE).

Context
- SIE is an OpenAI-style inference API. Python SDK: `from sie_sdk import SIEClient`; TypeScript: `@superlinked/sie-sdk`.
- Base URL: https://api.superlinked.com (or my regional endpoint). Auth: Bearer key from env `SIE_API_KEY` (never hard-code it).
- Model: opensearch-project/opensearch-neural-sparse-encoding-doc-v3-gte (SIE primitive: /encode). Keep the model id configurable.

Task
- Input: a search query plus the listings to rank.
- Behaviour: return the query and listing sparse vectors, with vocabulary ids and weights
- Send two POST /v1/encode requests: one item {text: query} with params {output_types: ["sparse"], options: {is_query: true}}, then all listings in one batch with params {output_types: ["sparse"], options: {is_query: false}}. Read items[i].sparse.indices (vocabulary ids) and .values (weights). Return these native sparse vectors without adding application scoring or ranking. In Python, use SIEClient.encode(..., output_types=["sparse"], is_query=True) for the query call and is_query=False for the listings; in TypeScript, pass isQuery: true and isQuery: false respectively. Do not issue one request per listing.

Deliverables
- A typed client wrapper, an application-level function for this task, error handling for timeouts/empty input, and unit tests with a stubbed client.
- Wire it into my existing stack (ask me which framework if unclear) and add a short usage example.

Run

Output

Search terms, IDF lookup

way 3.49##fa 5.74##ir 6.48comfort 6.35##ers 4.19

ListingScore

wayfair basics® comforter setExact 13.87
wayfair sleep medium weight all season polyester down alternative comforterExact 13.28
wayfair basics® reversible quilt setExact 12.47
wayfair basics single curtain rodIrrelevant 6.22
wayfair basics peva shower curtain linerIrrelevant 6.20
wayfair basics round single curtain rodIrrelevant 6.14

The API returns sparse vectors. This display ranks the recorded vectors by dot product; labels are Wayfair's.

Encode your catalog once; searches need no model call

Serverless API

We host the models; you call one API.

No idle GPUs, pay for what you use
Fits your stack: SDK, API, CLI, MCP
Self-host the same open-source engine
SOC 2 Type 2, US or EU data residency
Create an account to try Serverless

Get started

Self-hosted (Kubernetes)

Your data stays in your cloud, backed by our engineers.

Terraform modules for your cloud
Apache-2.0, same engine as Serverless
GPU workers scale to zero between jobs
Separate worker pools for each tenant

Agent prompt Deploy SIE to our AWS account with the superlinked/sie/aws Terraform module. Docs: superlinked.com/docs/deployment.

Deploy SIE to our Google Cloud project with the superlinked/sie/google Terraform module. Docs: superlinked.com/docs/deployment.

Deploy SIE to our Azure subscription with the superlinked/sie/azure Terraform module. Docs: superlinked.com/docs/deployment.

Deploy SIE to our Alibaba Cloud account with the superlinked/sie/alicloud Terraform module. Docs: superlinked.com/docs/deployment.

Talk to us Deploy guide

Run locally

The open-source server on your own machine.

NVIDIA
One container, no cluster
Process your data fully offline
Same SDK and model IDs as Serverless

docker run --gpus all -p 8080:8080 ghcr.io/superlinked/sie-server:latest-cuda12-default

Quickstart

Switch to SIE in 5 minutes

from elasticsearch import Elasticsearch

client = Elasticsearch("https://your-elasticsearch-endpoint", api_key="...")
model = ".elser-2-elastic"

query_vector = client.inference.sparse_embedding(
    inference_id=model, input=query,
)
document_vector = client.inference.sparse_embedding(
    inference_id=model, input=document,
)

from pinecone import Pinecone

pc = Pinecone(api_key="...")
model = "pinecone-sparse-english-v0"

query_vector = pc.inference.embed(
    model=model, inputs=[query], parameters={"input_type": "query"},
)
document_vector = pc.inference.embed(
    model=model, inputs=[document], parameters={"input_type": "passage"},
)

SIE

from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")
model = "opensearch-project/opensearch-neural-sparse-encoding-doc-v3-gte"

query_vector = client.encode(
    model, [{"text": query}], output_types=["sparse"], is_query=True,
)
document_vector = client.encode(
    model, [{"text": document}], output_types=["sparse"],
)

from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")
model = "opensearch-project/opensearch-neural-sparse-encoding-doc-v3-gte"

query_vector = client.encode(
    model, [{"text": query}], output_types=["sparse"], is_query=True,
)
document_vector = client.encode(
    model, [{"text": document}], output_types=["sparse"],
)