TNG Shopper helps large multi location retailers grow by generating a custom website per product per store in real time. Retailers provide store and product data, and TNG uses AI to enrich that data into full product experiences.
They partnered with Empromptu to build an automated enrichment system that could handle thousands of locations, multiple languages, and large daily volumes without expanding the engineering team or rewriting their platform.
The challenge
TNG already had a working system, but it could not support the scale of their growth. In one week they signed four large retail enterprises that represented 1,600 store locations at full scale. Together with 73 existing active sites, they were rapidly approaching 190 locations, each with thousands of products that needed enriched content.
The team was, in their own words, throttled by the number of customers they could take on. They could not accept more demand without a better way to generate high quality, localized data for every product and store.
Throttled by size and scale of retail data
Every location required its own version of product data. The same t shirt in a family oriented Madrid neighborhood needed different messaging than the version sold in a college town. Product descriptions, FAQs, use cases, and feature highlights all had to be tailored to demographics, local events, weather patterns, and cultural context.
This level of localization at scale created a bottleneck. The business opportunity was there, but the content pipeline could not keep up.
The analysis paralysis of AI implementation
The team needed to process 50,000 product requests every day, each going through multistep AI optimization, transformation, and formatting. They considered building their own orchestration pipeline, but knew this would take roughly half a year to implement.
CTO Jason Schwartzman, with over twenty years in tech and a decade of building ML models, understood that they needed multi-language support at a high level, production grade reliability, and strong contextual understanding. Building it all internally would mean pulling engineers away from the core product, which was not an option.
The vision vs reality gap
CEO Daniel Manzela had a clear vision: move fast, deliver high quality work, and keep the entire company moving without waiting on code. He wanted tools that let the team execute quickly without requiring deep AI expertise.
He described a future where developers rely less on hand written code and more on platforms like Empromptu to deliver results faster, as long as the system works in production, meets SOC and compliance requirements, and delivers strong outcomes.
The missing piece was the infrastructure to bridge that vision with day to day reality.
The solution
TNG engaged Empromptu to build an enterprise grade Automated Product Content Enrichment System. The goal was a complete AI application that plugged into their existing stack, handled enrichment at scale, and did not require additional headcount or a platform rewrite.
Empromptu delivered the agentic infrastructure, LLM ops tooling, and evaluation frameworks so TNG could focus on the retail use case instead of building orchestration from scratch.
What TNG Shopper is building
Multi language data enrichment pipeline
The system takes product and location data and generates:
- •Demographically aware customer personas for each product location combination
- •Localized use cases based on city demographics, weather, local events, and cultural context
- •Context specific product features that surface the most relevant benefits per location
- •Frequently asked questions tuned to local search patterns and concerns
- •Seasonal content that adapts to local trends
Human in the loop optimisation
Quality is maintained through built in feedback mechanisms:
- •Sample based annotation for reviewing outputs
- •Manual scoring interfaces that feed back into training and refinement
- •Automated optimization based on patterns in human feedback
This brings a traditional data science tuning loop directly into the production workflow.
Production scale infrastructure
The platform is designed for real world scale:
- •50,000 daily product processing requests
- •Support for English, Spanish, Hebrew, and expansion toward 14 plus languages
- •JSON based ingestion and output for clean API integration
- •SOC 2 aligned infrastructure with GDPR needs addressed
- •Architecture capable of supporting thousands of locations concurrently
Why Empromptu
Empromptu allowed TNG to avoid building a complex orchestrator and AI pipeline themselves. As Jason put it, the agentic approach removed the need for a half year internal build.
The platform provided strong multi language capabilities out of the box, using models with better contextual understanding than the alternatives TNG evaluated. It was built for real business use cases, not just prototypes, and shipped with LLM ops, evaluation, and automatic optimization after repeated runs.
On top of the technology, TNG received a hands on partnership: dedicated Slack channels, onboarding for the whole team, and forward deployed engineers when needed, backed by more than fifteen years of AI and dev tools experience.
The vision: AI first retail intelligence at global scale
TNG is moving from a service model constrained by manual work to a platform model powered by AI. With Empromptu running the enrichment pipeline, they can:
- •Accept enterprise clients without worrying about capacity
- •Scale from hundreds of locations to thousands without matching headcount
- •Maintain data quality and localisation while expanding globally
- •Keep engineers focused on core product innovation instead of infrastructure
They are building the infrastructure that will define how AI driven retail experiences operate across markets and languages.
Key takeaways
Scaling without a specialized AI team
TNG is building enterprise grade AI infrastructure without hiring ML specialists or committing to a long internal build.
Multi language production quality
The system supports English, Spanish, Hebrew, and more, with production level quality rather than prototype level outputs.
50,000 daily requests without an overhaul
High volume processing is handled without rewriting existing systems or pausing the core roadmap.
Human in the loop at scale
Integrated feedback flows keep quality high while still gaining the leverage of automation.
Building your own AI features
TNG Shopper shows how a retail technology company can handle production scale AI workloads without expanding the team or investing months into custom orchestration.
If you are building AI powered features for your own platform, Empromptu provides the same foundation: agentic builders, production ready infrastructure, and hands on support so you can focus on the product and not the plumbing.
