How Alice Helped One of the Most Recognized Toy Companies Gain Full Control of Its Child-Facing Product

The client had built a personalized storytelling experience for children. Users upload a photo and share their interests; the application generates a custom story featuring them alongside a beloved character, complete with AI-generated imagery, available to purchase as a physical product.

To enforce content policy across user inputs and AI-generated outputs, the client had been relying on the out-of-the-box guardrails that came with the LLM they were building on. When that LLM made an unannounced change to its guardrails, the effect rippled through the client's entire production environment.

Depending on default guardrails meant that enforcement rules could change at any time, for any reason, without warning. The client came to Alice seeking a solution that offered greater control and customization across both text and image generation, within the three-week window they had before going to production.

There was also a challenge specific to their product format that required a high degree of nuance. Characters, settings, costumes, and props are expected to reflect their narrative context accurately, and meeting this requirement demanded contextual reasoning that standard out-of-the-box guardrails are not built to provide.

Alice delivered a complete guardrails solution covering text and image enforcement end-to-end, within two weeks.

Four custom detection capabilities were built for the client's deployment:

→ Competitor logo detection, identifying brand marks in user-uploaded and AI-generated images

→ Children in risky contexts, detecting unsafe imagery involving minors

→ Age-bucket classification, with a specific enforcement threshold at 13 and above

→ Narrative context detection, an LLM-based capability that flags contextually inaccurate content in generated images

Because the client had no production data at the start of the engagement, Alice ran sample data through the platform and provided a full statistical analysis comparing out-of-the-box guardrail performance against Alice's custom models. The result gave the client a clear, evidence-based view of what the customized solution delivered, and the ability to iterate on their actual content before going live.

Four custom detection capabilities replaced a generic moderation configuration that had already proven it could change without notice. Enforcement was now built to the client's policies, characters, and child safety requirements, not borrowed from a platform default.

Custom enforcement delivered measurably better precision on the risks that matter for a child-facing product. Product and safety stakeholders had the evidence they needed to move forward. The client launched with a guardrails layer that reflects who they are and what their product demands, and, for the first time, full control over their own safety layer.