phoenix
phoenixPhoenix - Your AI Assistant
Inactive session will be ended in 24 hours
End Session

Product Data enrichment

  • Posted:
  • Proposals: 25
  • Remote
  • #4511145
  • Open for Proposals
Óscar P.MD Rakibul I.Joshua A.Digital-Elites T.Mahmood U. + 20 others have already sent a proposal.
  • 9

Description

Experience Level: Expert
I have 12,000 products from a website, I need to enrich this data with lots more data points and specifications with data from online to create a wide database which can be used for analysis, it will require multiple data points, manufacturers websites, retail sites.

Not all products use the same SKU's on different sites so LLM will need to be used to get matches where possible. It will not be possible to get 100% complete data but we need to get it much higher than the current data set.

This needs completing within 10-14 days

AI will be a big part of gathering this data from multiple online sources.

This is within Airsoft so for example an Airsoft Rifle would need
Platform
Length
barrel length
power source
gearbox version
stock version
trigger
weight
fps

and more

I have a full set of data points which are required already they just need the data obtaining

New Proposal

Login to your account and send a proposal now to get this project.

Log in

Clarification Board Ask a Question

  • Muhammad Atif K.

    What percentage of your 12,000 products have a globally unique identifier (EAN/UPC/GTIN/MPN), and what percentage only have retailer-specific SKUs?

    When multiple retailers disagree on a specification (e.g., FPS or weight), what is your source-of-truth hierarchy? Manufacturer > Distributor > Retailer > Community?

    How do you want confidence represented? Should every enriched field include confidence score, source URL, extraction timestamp, and extraction method?

    Should entity resolution be deterministic first (SKU, GTIN, MPN, manufacturer) and only invoke LLMs when deterministic matching fails, or should LLM semantic matching always participate?

    How much false-positive matching is acceptable? For enrichment pipelines I usually optimize for >99% precision even if recall drops slightly because incorrect specifications poison downstream analytics. Is that your preference?

  • Mahmood U.

    Could you let me know what format the current product dataset is in (CSV, Excel, database export, or another format)?