Before

The company appears to be a solo-built project named "Before", self-described as a safety layer for AI interactions that pauses risky actions (e.g., BUY, PAY, SEND) and provides clear verdicts (GO / FIX / PAUSE). It is positioned as a tool to help users make safer digital choices by offering explanations and control over AI-assisted decisions. The project was submitted to the OpenAI 2026 hackathon.

The most important open question is: What is the actual commercial viability or scalability of this concept, given that it is currently presented as a prototype built in a hackathon setting with no evidence of revenue, customers, or product-market fit beyond the author’s own claims?

• The description states that BEFORE is a safety layer for AI interactions.
• It detects dangerous intent patterns such as BUY, PAY, SEND, CLICK, SIGN, and POST.
• It runs a six-check pipeline and returns one of three verdicts: GO / FIX / PAUSE.
• It provides explanations of what was checked, what was missing, and what safer step to take next.
• The system is described as policy-driven, with model analysis, deterministic validation, domain checks, and server-side enforcement.
• It was built in a hackathon setting (Devpost submission), not as a production-ready product.

Note: No evidence of actual product functionality or user experience beyond the author’s description.

• The project is positioned to address the problem of people trusting AI too much in high-risk moments.
• It claims to give users control back by making risky actions pause, explain themselves, and ask for confirmation.
• The positioning evolves from a general safety concern (AI misuse) to a specific tool that makes digital decisions safer through transparency and control.
• The author states that BEFORE is not about blocking everything but about making every risky action pause and explain itself.

Inference: The positioning reflects a shift from AI-as-risk to AI-as-trustable assistant, based on the author’s own claims.

• The description does not name specific customer segments or personas.
• It implies that users who are in a hurry and trust AI-assisted actions (e.g., drafting messages, making payments) are the target.
• The product is described as useful for everyday copilots across domains like banking, shopping, scheduling, document signing, and account actions.

Not evidenced: No explicit ICP or customer segmentation beyond vague descriptions of user behavior.

• No pricing information, monetization strategy, or business model is provided.
• The description does not mention any revenue streams, subscriptions, or commercial partnerships.
• It is described as a tool that could be integrated into other platforms or used standalone, but no commercial structure is outlined.

Not evidenced: No evidence of how the product would generate value or income.

• Built with: codex, gpt-5.6sol-ultra, gstack, openai, typescript.
• The system uses a policy-driven pipeline including model analysis, deterministic validation, domain checks, and server-side enforcement.
• It includes evaluator tests for various cases (clear, concern, unknown, adversarial) to ensure consistent behavior.
• UX was tuned to avoid annoyance while maintaining clarity.
• Challenges included avoiding “fake confidence” and pacing the UX appropriately.

Inference: The technical stack suggests a strong AI integration with a focus on safety and user experience tuning.

• The project was submitted to the OpenAI 2026 hackathon.
• It is described as a working flow, but no evidence of real-world usage or adoption.
• No mention of users, customers, or any form of traction beyond the author’s own account.
• The product is described as a prototype built in a short timeframe (a buildathon), not a mature product.

Not evidenced: No data on usage, retention, or product maturity beyond self-reporting.

• The description does not mention competitors or similar tools.
• It positions itself as a safety layer for AI interactions, which could overlap with AI governance, digital trust platforms, or cybersecurity tools.
• No evidence of existing solutions in the market or competitive differentiation is provided.

Not evidenced: No competitive landscape or positioning relative to other tools.

• The project is described as a hackathon prototype with no commercial traction or product-market fit.
• It lacks any evidence of revenue, customers, or real-world usage.
• The author’s own claims are unverified and self-reported.
• No clear path to monetization or scalability is evident.
• The system relies heavily on AI models and deterministic validation — a potential risk if model performance degrades or becomes unreliable.

Inference: The lack of evidence for traction, revenue, or product-market fit raises significant commercial viability concerns.

1. What is the actual user behavior you’ve observed in prototype testing?
2. How do you plan to scale this from a hackathon prototype to a usable product?
3. Have you tested the UX with real users, and how did they respond?
4. What are your plans for monetization or commercial partnerships?
5. How do you intend to handle edge cases where model confidence is low but deterministic checks are insufficient?