RoboYogi Onboard

Inspiration

Guardian's product onboarding is manual because supplier documents, certificates, images, and raw product data arrive in inconsistent formats and someone has to read, check, and re-key all of it before a product can go live. This is the exact problem RoboYogi was born to solve — our marketplace's supplier verification agent already does document and claim verification for conscious-living products. Guardian's problem statement is our core engine applied to health and beauty retail at chain scale.

What it does

RoboYogi Onboard is an AI product onboarding and compliance agent for Guardian. It ingests supplier submissions — registration documents, product certificates, ingredient lists, product images, and raw product data — extracts the structured fields Guardian's catalog needs, checks certificates and compliance claims against requirements, flags gaps or inconsistencies for human review, and outputs onboarding-ready product records. Manual weeks become hours, with humans reviewing exceptions instead of re-keying everything.

How we built it

The agent pipeline extends RoboYogi's supplier verification engine. A document-extraction agent parses supplier files (PDFs, images, spreadsheets) into structured product data; a compliance agent validates certificates and claims against Guardian's category requirements, checking expiry dates, issuing bodies, and claim-certificate consistency; and a review agent packages everything into an onboarding record with a confidence score and a flagged-exceptions list. Nothing auto-publishes — every record lands in a human review queue ordered by confidence, so the team's time goes to the genuinely ambiguous cases.

Challenges we ran into

Certificate verification credibility. Extracting text from a certificate is easy; knowing whether the certificate actually supports the product claim is the hard part. We had to build explicit claim-to-evidence matching rather than letting the model wave a document through because it "looks like" a certificate. Handling inconsistent supplier formats without silently dropping fields was the other grind — a wrong blank is worse than a flagged gap.

Accomplishments that we're proud of

The compliance agent never overclaims: when evidence is missing or ambiguous, it flags rather than guesses. Building that restraint in as a hard rule — not a tuning preference — is what makes the output trustworthy enough for a retailer to act on.

What we learned

Onboarding automation isn't about replacing the review team — it's about inverting their workload. When the agent handles extraction and first-pass verification, humans stop being data-entry clerks and become exception judges. That's where the real time saving lives.

What's next for RoboYogi Onboard

Extending compliance rule sets per product category, and connecting verified onboarding records directly into catalog and pricing systems so a passed review goes live without a second manual step.

What we learned

Built With

• node.js
• openai