Published

How a Celebrity Stylist Sourced Products in Seconds

A celebrity stylist's team was burning hours per request hunting products on Google. A GPT-4o Vision pipeline now extracts style cues from images and returns matches in seconds.

MVP in 5 weeks

Delivered from kickoff to launch

< 4 weeks

Implementation Time

Not disclosed

Project Cost

the challenge

Sourced By — a luxury product sourcing business built around celebrity stylist Gab Waller — was scaling faster than its research process could support. Each sourcing request required hours of manual Google searches and image-by-image review to find products matching client specifications. There was no way to increase request volume without proportionally increasing the time Gab and her team spent on manual research.

what they built

Headstart built a custom AI vision pipeline for Sourced By that replaced manual search with an image-recognition and product-matching system. GPT-4o Vision API processed product images to extract style attributes, materials, and category tags; Claude handled natural language query interpretation and matching logic. The system returned structured product recommendations in seconds rather than hours. Built in 5 weeks as an MVP, the pipeline also generated a proprietary training dataset from Sourced By's historical sourcing data — a defensible asset the business now owns.

Headstart designed a two-model pipeline that separates image understanding from query matching. GPT-4o Vision API processes incoming product images, extracting structured attributes including materials, silhouette characteristics, aesthetic categories, and luxury-specific style markers — the nuanced vocabulary that matters in high-end sourcing. Claude then interprets the natural language sourcing query and applies the matching logic, returning ranked product recommendations in structured form. The matching system was trained on Sourced By's historical sourcing data — past requests, approved products, and client preferences — creating a proprietary dataset the business now owns as a defensible asset. Building this matching logic to perform reliably on luxury-specific attributes within the MVP timeline was the key technical challenge: off-the-shelf models trained on general product search lack the precision required when a client distinguishes between ivory silk charmeuse and cream satin. The full pipeline — from kickoff to working MVP — was delivered in five weeks.

best fit for

Luxury e-commerce and personal shopping businesses scaling research-intensive operations; AI agencies building vision-enabled product matching tools.

Ai ROLE

GPT-4o Vision API processes incoming product images to extract structured style attributes, material properties, and category tags. Claude interprets natural language sourcing queries from the client team and applies matching logic to connect those queries against the extracted product attribute data. Together, the two models replace the manual image-by-image review process, returning structured product recommendations in seconds.

impact

MVP in 5 Weeks

Full AI vision pipeline — image recognition, product matching, and recommendation output — delivered from kickoff to working MVP in 5 weeks.

Manual Research Eliminated

Hours of per-request Google search and image review replaced by automated AI pipeline delivering structured results in seconds.

Proprietary Training Dataset Created

Historical sourcing data converted into a labeled training dataset — a proprietary AI asset owned by Sourced By for future model fine-tuning.

implementation complexity

The solution required building a custom multi-model vision pipeline integrating GPT-4o Vision API and Claude, plus developing product attribute extraction logic and a matching system trained on proprietary historical data. This goes beyond off-the-shelf tooling but stops short of full custom model training.

Nicole Hedley

Founder & CEO @ Headstart

Headstart

Founder and CEO of Headstart, an AI-native company transforming businesses by rapidly building innovative AI solutions that dramatically accelerate product development.

Get an intro

Talk to this team

industry

Retail & E-Commerce

business organization

Operations

Supply Chain & Procurement

AI TYpe

Computer Vision

Recommendation Systems

value type

Time Savings

Revenue Growth

Headcount Avoidance

other(firm) Case studies

frequently asked questions

How did a small retail and e-commerce business source products in seconds with computer vision?

The experts built a two-model pipeline that separates image understanding from query matching: GPT-4o Vision extracts structured style attributes from product images, then a second model interprets the natural-language sourcing query and returns ranked matches, trained on the business's historical sourcing data. Searches that took hours of manual Google review now return results in seconds.

What AI tools and models were used in this product-sourcing project?

The pipeline used the GPT-4o Vision API to extract materials, silhouette, and luxury-specific style markers from images, and Claude to interpret the sourcing query and apply matching logic, with ChatGPT also in the toolset. The approach combined computer vision and recommendation systems.

What results did the business achieve?

Three outcomes: hours of per-request manual search and image review were replaced by an AI pipeline returning structured results in seconds; historical sourcing data was converted into a proprietary, owned training dataset; and the full vision pipeline was delivered from kickoff to working MVP in five weeks.

How long did the AI vision pipeline take to build?

The full pipeline — image recognition, product matching, and recommendation output — was delivered from kickoff to working MVP in about five weeks.

Who is this computer vision sourcing approach best for?

Luxury e-commerce and personal shopping businesses scaling research-intensive operations, and AI agencies building vision-enabled product matching tools.

Have a similar challenge?

Ask whether this would work for you, or describe what you're trying to solve.

TELL US WHAT YOU'RE EXPLORING