By Teresa Torres · producttalk.org · @ttorres on X · LinkedIn
This episode of Just Now Possible features a conversation with Priya Nakra and Olivia Rowley of Override Labs, a nonprofit building technology to prevent gender-based violence. Their flagship product, Is This Okay?, gives teenage boys a private, judgment-free space to reflect on ambiguous sexual scenarios, using AI guidance grounded in clinical research and motivational interviewing. The guests describe how they validated demand, designed a deterministic risk classification layer that runs before the language model is invoked, and avoided any response that could be read as permission to cause harm. They also discuss a "South star" framing centered on the worst-case outcome, privacy-by-design choices such as no accounts or cross-session tracking, and the difficulty of measuring prevention when success means something did not happen. The discussion is a useful case study in purpose-built AI product work where the goal is prevention rather than scale.
01Key takeaways
- Define the worst-case outcome as the guiding constraint, then design the product explicitly to avoid it.
- Run deterministic risk classification before invoking a language model, and let the tier shape the response.
- Ground AI tone and structure in clinical expertise and established frameworks such as motivational interviewing.
- Treat the absence of data collection as a product feature when user trust is central to the mission.
- Validate demand with existing public conversations before committing to a narrow audience.
- Measuring prevention requires proxy metrics, since success often means something did not happen.
02Key sections
- Why the nonprofit exists
- The founders explain their motivation for leaving a long tech career to address gender-based violence prevention. They frame the mission around building tools that reduce harm rather than maximize engagement.
- Validating the problem and narrowing scope
- The team tested demand by analyzing public discussion and narrowed the focus to teen consent questions. Scoping to a specific, under-served audience made the product tractable.
- Clinical grounding and design choices
- A licensed therapist and coaches shaped the tone, eval rubric, and response structure. Choices such as avoiding any "green light" answer were made to prevent the tool from being used to justify harm.
- Architecture: risk tiers before the model
- Risk classification runs deterministically before the language model is called, and responses are tailored by tier. A three-part, motivational-interviewing-based structure guides the conversation toward reflection.
- Privacy, measurement, and ecosystem
- Privacy by design means no accounts or cookies, which the team treats as a feature. They discuss measuring prevention indirectly and a broader product ecosystem including a web game funnel and future institutional access.
03From the post
“Listen to this episode on: Spotify | Apple Podcasts What if AI could help prevent sexual assault before it happens — without tracking users, judging them, or handing them a verdict? In this episode of Just Now Possible, Teresa Torres talks with Priya Nakra (Founder and Product Lead) and Olivia Rowley (AI”
04Frameworks mentioned
Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.