Gazelle Trace

What the company appears to be: Gazelle Trace is a teacher-bounded adaptive math tutor for elementary learners, built as a hackathon project. The product uses GPT-5.6 and other AI tools to diagnose student misconceptions and propose Socratic next steps, while enforcing strict safety and approval gates. It is designed to operate within constraints set by teachers, with a focus on transparency and control.

What changed: This is a self-reported prototype built in one week for the OpenAI 2026 hackathon. The author states that it was not intended as a full product but rather as a proof-of-concept demonstrating how AI can be used safely within educational settings, with explicit teacher oversight.

Single most important open question: Is there evidence of traction or commercial viability beyond this prototype? The description does not indicate any revenue, customers, or adoption beyond the hackathon submission.

The description states that Gazelle Trace is a teacher-bounded adaptive math tutor for elementary learners. It allows students to submit typed or handwritten work, which is then processed by GPT-5.6 to identify observable misconceptions and propose one short Socratic next step without revealing the stored answer.

A separate verification pass and deterministic policy gate decide whether that move may reach the learner. The system exposes these decisions as a human-readable trust trace, including input moderation, teacher-approved curriculum evidence, answer-leakage detection, output moderation, and independent verification.

The teacher can approve or hold the move, and a second learner response reruns the complete pipeline before the tutor adapts difficulty.

Evidence:

• The product is described as a math tutor for elementary learners.
• It uses GPT-5.6 for diagnosis and verification.
• It includes a trust trace mechanism that shows decision-making steps.
• It enforces teacher approval and deterministic policies to prevent unsafe or inappropriate outputs.

Inference:

• The system appears to be designed with safety and transparency in mind, particularly around student privacy and teacher control.
• It is not described as a full learning management system but rather a specific tool for tutoring within constraints.

The description states that Gazelle Trace addresses the problem of AI tutoring failing when fluency is mistaken for trust. It aims to avoid replacing teachers or rebuilding an entire LMS, instead focusing on a narrow but critical issue: how AI tutors make decisions and whether those decisions are safe and appropriate.

It positions itself as a solution where:

• A hint can sound helpful while being ungrounded, too advanced, privacy-invasive, or so complete that it removes the learning.
• Teachers need to inspect why a tutor chose its next move and retain control over whether it advances.

The author notes that most tutor demos end when the model generates a plausible answer. Gazelle treats generation as the beginning of the decision process, with visible trust traces, separate verifier authority, deterministic policies, teacher checkpoints, two-turn adaptation, and runnable judge challenge sets.

Evidence:

• The product is framed as addressing a specific gap in AI tutoring — lack of transparency and control.
• It emphasizes that it does not aim to replace teachers or build a full LMS.
• It introduces a novel approach: treating generation as the start of decision-making rather than an endpoint.

Inference:

• This positioning reflects an attempt to differentiate from typical AI tutoring tools by emphasizing safety, teacher control, and explainability.
• The focus on testable claims suggests a desire for credibility in a competitive hackathon environment.

The description states that Gazelle Trace is built for elementary learners and their teachers. It is described as a math tutor, and the system is designed to be used in educational settings where teacher oversight is critical.

It does not specify whether it targets individual schools, districts, or platforms, nor does it describe any segmentation strategy beyond grade level (K–6).

Evidence:

• The target audience is elementary learners.
• Teachers are explicitly mentioned as needing to inspect decisions and retain control.
• The system is built for K–6 math education.

Inference:

• The product may be aimed at educators who want more control over AI-assisted learning tools.
• It could potentially appeal to schools or edtech platforms looking for safer, teacher-bounded tutoring solutions.

There is no evidence in the description of a business model or pricing structure. The project is described as a hackathon submission and does not mention any monetization strategy, subscription plans, or customer acquisition methods.

Evidence:

• No mention of revenue streams.
• No indication of pricing or licensing models.
• No reference to enterprise sales or B2B customers.

Inference:

• The product is likely not yet commercialized.
• If it were to be monetized, it would probably involve a SaaS model for schools or teachers, but this is speculative.

The project was built using:

• Cloudflare Workers
• Codex and GPT-5.6
• Next.js / Vinext
• OpenAI Responses API
• Omni-moderation-latest
• TypeScript
• Zod

Key technical features include:

• A deterministic PII screen to prevent contact information from reaching models.
• Use of omni-moderation-latest for learner text and images.
• Retrieval only of teacher-approved lesson evidence.
• Schema-validated structured diagnosis via GPT-5.6.
• Deterministic code rejecting stored-answer variants in prompts.
• Independent verification pass by GPT-5.6 to approve or reject evidence support, grounding, age level, and non-leakage.
• Only fully validated candidates reach the learner; others become teacher handoffs.

Evidence:

• The system uses multiple AI tools and frameworks.
• It includes safety mechanisms like PII screening and moderation layers.
• It enforces deterministic policies to override model decisions.
• It is packaged as a self-contained Cloudflare Worker-compatible artifact.

Inference:

• The architecture suggests a strong emphasis on safety and compliance.
• The use of multiple AI models and tools indicates complexity in implementation.
• The modular nature of the system implies potential for scalability or integration into larger platforms.

The description states that Gazelle Trace is a hackathon prototype built in one week. It includes:

• A public two-turn product.
• Multimodal GPT-5.6 diagnosis and independent verification.
• A bundled handwritten sample for judge testing.
• Twelve-case live text-and-image release matrix covering various scenarios.
• Packaging as a Cloudflare Worker-compatible artifact.

It also notes that the next step is educator validation, including measuring teacher agreement on diagnosis and next-move quality, studying false holds, and defining provider-retention controls.

Evidence:

• The product was submitted to the OpenAI 2026 hackathon.
• It includes a live demo with test cases.
• It has no revenue or customer data beyond the prototype.
• No mention of user adoption or feedback from real users.

Inference:

• The project is early-stage and experimental.
• There is no indication of traction, usage metrics, or market validation.
• The author explicitly states that it does not claim COPPA compliance or measured learning gains.

The description does not provide any information about competitors or the broader competitive landscape. It focuses on what Gazelle Trace does differently rather than comparing itself to existing products.

Evidence:

• No mention of competitors.
• No reference to similar tools in the market.
• The author emphasizes that this is a narrow, focused solution compared to general-purpose AI tutors.

Inference:

• The competitive context is unknown.
• It may compete with other AI tutoring platforms or edtech tools, but no such comparison is made.

Several risks and red flags are evident from the description:

1. Prototype-only status: The product is described as a hackathon submission with no commercial traction or real-world deployment.
2. No revenue or customer data: There is no evidence of monetization, users, or adoption beyond the prototype.
3. Limited scope: It only covers K–6 math and does not claim to be scalable across subjects or grade levels.
4. Founder-only team: The project was built by a single person (Laith Askar), which raises questions about scalability and long-term development.
5. Unverified claims: All descriptions are self-reported and unverified; there is no independent validation of performance or safety.

Evidence:

• The product is described as a one-week hackathon effort.
• No mention of any real-world usage or feedback.
• No indication of team size beyond one person.
• No evidence of product-market fit or scalability.

Inference:

• The risk of failure in transitioning from prototype to product is high without further development and validation.
• Lack of traction makes it difficult to assess commercial viability or market demand.

1. What specific feedback have you received from educators or teachers who tested the prototype?
2. How do you plan to scale beyond a single-person development team?
3. Are there any plans to integrate with existing educational platforms or LMS systems?
4. What are your thoughts on expanding beyond K–6 math into other subjects or grade levels?
5. How do you intend to validate the safety and effectiveness of the system at scale?
6. Have you considered how this product might be monetized, and what pricing model would work best for educators?