Welcome
Last updated
Last updated
The Product LLM is an API service that provides large language model workflows specialized in e-commerce product matching.
We serve service providers who need to embed product intelligence into their e-commerce applications.
🔎 Product Search
🧠 Product Matching
With The Product LLM, you can access high accuracy product matching models with costs designed for high volume usage. As we find cost savings, we pass on those savings to you, without the headaches of deploying your own models.
theproductllm-mini
(certainty = high)
97%
61%
$0.26 / M Tokens
theproductllm-mini
(all results)
90%
89%
$0.26 / M Tokens
Accuracy and retrieval metrics are based on our test data of difficult-to-score product matches. These are ~50% difficult, borderline test cases.
We train our models on data generated from 100,000+ product match candidates, selecting tens of thousands of difficult edge cases from a broad product categories and ecommerce sites. All training data must pass human review with multiple reviewers before being used to train our models.
Every successful match comes with a certainty score. Users can distinguish between high certainty matches and medium certainty matches.
standard, double
65%
$0.001
Product retrieval reflects the percent of known matches that were found in the top 5 search candidates. At 65%, we are finding approximately 2 out of 3 products. To achieve this, we perform multiple search queries, with AI query bots eliminating bad candidates, and generating additional web search queries.
We offer 50 search queries per second for high volume customers, and if you need more, you can contact us to increase your search volume as needed.
We have two pipelines for searching and matching products.
However, many service providers have access to APIs for the websites their interested in - such as Amazon, Walmart, and eBay. Therefore, if you have access to page data via API, you can get 10x cheaper costs by bringing your own data.
This cheaper, multi-step process would involve the following steps:
Search a product with our /search-products endpoint, and we'll give you the top candidates, after eliminating definite non-matches
Ping an external API to get detailed product descriptions
Score the match with /match-products, using product data from step 2
If you bring your own data, a standard search and match workflow will cost between $0.0012-$0.0016 to search 1 product and evaluate 5 matches.
If you use our end-to-end /search-products-with-page-visit, a single Tier 1 website page visit will cost $0.005 per page visit (e.g., Amazon), plus the $0.0012 - $0.0016 from searching and matching.
One of the best strategies to employ if your use case permits it, is to filter out candidates you're not interested in before requesting a page visit. For example, if you're only interested in profitable products, you should filter out unprofitable candidates before visiting web pages to evaluate the match quality.
standard, double
Bring-your-own-data
$0.0012 - $0.0016
standard, double
/search-products-with-page-visit
$0.006 - $0.025
Note on ambiguous data: Please note that vague input data that does not distinguish between two similar products can be more difficult to score, and accuracy could be lower in these cases, especially when the match certainty is "medium." Luckily, the certainty score helps solve this problem, since users can choose whether to filter out matches with lower certainty. See .
The expensive, simplest way is to query /, where we handle everything - search, page visits, and matching against page visit data. That costs $0.001 plus $0.005 per page visit we have to make.
🏃♂️Quickstart Guide
🔎 /search-products
🧠 /match-products