Skip to content

Wegmans Marketplace Validation Agents

Every promoted item is checked against the source of truth, with problems surfacing in minutes, not weeks.

Validates every promoted SKU across marketplace platforms against the source of truth, on demand.

What the agent does

Drives a browser agent for each marketplace to check promoted prices against the source of truth, and reports the mismatches.

Contributed by
MACH Alliance (Hub operator)
Industry
Retail, Grocery
Audience
B2C
01
The problem

Why this exists

Wegmans manages digital commerce across multiple channels and marketplace platforms that the company does not own or control. Product information, pricing, imagery, and promotional details on those marketplaces have to match the source-of-truth that lives inside Wegmans systems. Today that validation runs manually: a team member opens a product on the marketplace, eyeballs the data against an internal record, and moves on.

Read more

Manual validation tops out at about 5 products per hour and consumes around 100 person-hours a month. Only a fraction of the catalog gets checked, and only after a customer complaint or a routine sample. When a marketplace silently changes a product image or pricing rule, Wegmans finds out from a customer or never finds out at all.

The hypothesis behind this project is that autonomous browser agents can validate every promoted SKU on every marketplace, on demand and on schedule, instead of relying on manual sampling.

02
Outcomes

What this demonstrates

  • Validation shifts from manual sampling to exhaustive, on-demand coverage of promoted SKUs.
  • Discrepancies in price, image, and promotional content surface to the responsible team within minutes, not weeks.
  • Image comparison uses perceptual hashing so near-identical photos do not trigger false positives the way a strict pixel diff would.
Read more
  • Validation runs cache by SKU so repeat checks are cheap and the system learns which products drift most often.
  • The team that used to validate products gets that time back and spends it on exceptions, not routine checks.
03
How it works

Architecture

A request enters the system through an n8n webhook, either triggered by a schedule or by a marketing or operations user asking a question in plain language. A small language-translation agent powered by Ollama and Gemma 4 turns the natural-language request into a structured validation job. The system pulls the source-of-truth product record from the Wegmans Product API and the Algolia catalog index, then dispatches the validation work to one or more browser agents.

Read more

Each browser agent runs inside an ephemeral Vercel Sandbox running headless Chrome. Playwright drives the page navigation. Hyperbrowser provides agentic browser control where the next click depends on what the previous page rendered. The agent scrapes the marketplace product page, captures images, pricing, and promotional copy, and posts the raw observations into the validation pipeline.

Image comparison uses Scaleflex's perceptual hashing service. Unlike a cryptographic hash, a perceptual hash produces similar fingerprints for visually similar images, so a marketplace that slightly recompressed a product photo will still register as a match. The comparison returns a similarity score; the system flags only images that fall below an operator-configurable threshold.

Validation runs land in a Neon Postgres database keyed by SKU, so repeat validations skip work that has already been done and the system maintains a history of how each product has drifted over time. A second small Ollama agent synthesizes the raw findings into a business-friendly report and posts the result back through n8n to Slack or to whichever surface initiated the request.

n8n flow diagram showing the validation orchestration: webhook trigger, language interpretation, browser agent dispatch, image comparison, result synthesis, and Slack delivery.

n8n flow diagram showing the validation orchestration: webhook trigger, language interpretation, browser agent dispatch, image comparison, result synthesis, and Slack delivery.
Tech stack
n8nVercel SandboxNeon PostgresPlaywrightOllama (Gemma 4)Scaleflex (pHash)Vercel AI GatewayHyperbrowserAlgoliaClaude
Next step

Take the code

This one is published in full. Read it, run it, fork it. Nothing here is behind a conversation.

Request an intro

This one is the MACH Alliance’s own work. Ask about validating promotions across marketplaces, and the people who built it will answer.

Your details go to a named contact on the MACH Alliance team that built this work and can answer your questions. We aim to reply within 2 business days. Your details are handled under the MACH Alliance privacy policy.

Not ready to talk to anyone? MACH Insider is the Alliance mailing list. New solutions, program openings, and the research behind them, once a month.

Keep reading

Related solutions

  • MACH Alliance logo

    Wholesale buyers place and change orders in plain language, and see the price they will actually pay.

    SynapseAI B2B Wholesale

    AI Exchange Hackathon
  • MACH Alliance logo

    In six weeks, Dr. Martens and Conscia built a prototype on-site shopping assistant for drmartens.com that could search, size, and add to cart.

    Dr. Martens On-Site Shopping Assistant

    AI Exchange Hackathon
Verification

Public code

The implementation is published in full, so anything on this page can be checked against the repository. Anything the repository does not show is not established here.

Status: Reference prototypeAdded May 26, 2026A listing is an editorial decision, not a MACH Alliance certification.

Built something worth showing?

Put your agentic solution in front of the enterprise leaders who use the hub to decide what to build next. Submissions are open to MACH Alliance Members, Collaborators, or Ambassadors.