Principal Product Catalog Enrichment Specialist

Every fashion retailer sells a black T-shirt. So how does an AI agent pick one?

It starts in a catalog, such as Google’s Merchant Center or Meta’s catalog, and it filters on structured attributes: neckline, length, fabric, fit, size. Products with thin data or the wrong words drop out silently. When we asked an agent to buy a tall black T-shirt, it missed products described as "long", and it never offered the popular 3-pack. That’s a data problem, and it decides who wins the sale.

Speedway is ACOMM’s platform for winning the agentic shopper. You’ll build the enrichment engine at its core, and the research that tells us what actually persuades agents to buy.

In your first 90 days: you’ll be onboarded to the Speedway product vision, and you’ll work across all of our core product. We’re moving quickly, and you’ll ship models and code from the start.

In your first year:

• Relationship data, linking bundles, multipacks, substitutes and accessories, so a literal request still finds the right product.
• Feeds tuned to each catalog. Google’s conversational attributes, Meta’s catalog and the next surface, each optimised for how that agent reads products.
• The research program. Run real shopping journeys on every agent surface, learn which attributes, prices and answers win, and prove the lift for each retailer.
• Keep it fast and affordable at catalog scale: distillation, batching, and knowing when not to call a large model.

Google Cloud: BigQuery and Gemini Enterprise Agent Platform. Gemini and other frontier models, including multimodal ones. Python and SQL. Product taxonomies (Google, Shopify, GS1). We’ll choose the rest of the tooling with you.

• Production attribute extraction. Shipped LLM or ML systems that extract and normalise product attributes at scale, text and images.
• Taxonomy and schema design. You know how product taxonomies work and how to design category-specific attribute schemas.
• Rigorous evaluation. You design evaluations people trust: labelled sets, people reviewing the uncertain cases, confidence calibration, and catching models making things up.
• Experimentation. You measure lift, not just accuracy, and can explain the results to a retailer.
• A degree in computer science, statistics, machine learning or a related field, or equivalent experience.

• E-commerce catalog or product data at a retailer or marketplace.
• Google Merchant Center and product feed specifications.
• Search relevance, ranking or knowledge graphs.
• Published research or open-source work on attribute extraction or catalog quality.

• Industry-leading work in a field that changes every month. No one has a playbook for what makes an agent choose a product. You’ll help write it.
• Your research ships. What you learn goes straight into the product and into retailers’ sales.
• Remote-first. Build with agents, work from anywhere.
• Big ideas come from anywhere. A small, human-centered team of designers, engineers and data scientists.

Two interviews, three at most.

1. Technical conversation with our CTO (60 minutes): a model or data system you’ve built, and how you evaluated it.
2. Working session with the team (90 minutes): we take a real messy product catalog and talk through how you’d enrich and evaluate it. No unpaid take-home project.
3. If needed, a final conversation with the founders. Then references.

Applications close 31 January 2027. Email careers [at] acomm [dot] ai with your résumé or LinkedIn, and a few lines about the hardest data-quality problem you’ve solved.

ACOMM is an equal opportunity employer. We welcome applicants of every background, and we’ll make reasonable adjustments to our process if you need them.