Whetstone

The description states:

• Whetstone uses Codex to generate candidate coding exercises.
• Each exercise includes a problem statement, reference solution, and test suite.
• A sandbox executes the solution against the tests.
• If it passes, the exercise is published; if not, it goes into a discard log with failure details.
• The discard log serves as part of the product’s output.

Inference The system appears to be a proof-of-concept or prototype built during a hackathon. It does not appear to have any commercial revenue, customer base, or operational history beyond its author's development.

The description states:

• The original inspiration was that language models often generate broken exercises without knowing it.
• The solution is to make the model "earn" the problem by running it through a sandbox.
• The author claims that this approach shifts the product from filtering out bad problems to guaranteeing good ones.

Inference The positioning evolved from an idea of “filtering” to one of “guaranteeing,” based on observed discard rates during development. This reflects a shift in narrative rather than a change in core functionality.

The description states:

• The primary user is an instructor who uses generated exercises.
• The product provides insight into why certain problems fail, via the discard log.

Inference The target customer is likely educators or curriculum designers working with coding instruction. However, no evidence of actual users or customers exists beyond the author’s own use case.

The description states:

• No pricing model is described.
• The product is presented as a tool for instructors to validate exercises.
• There is no mention of monetization, subscriptions, or sales channels.

Not evidenced No commercial business model or pricing information is provided. The project appears to be a prototype without any indication of how it would generate revenue.

The description states:

• Built with Django and PostgreSQL for the generation loop and storage.
• Flutter Web renders the UI showing counts, logs, and exercises.
• Docker sandbox enforces security constraints (no network, read-only FS, timeouts).
• Codex and GPT-5.6 are used for generating candidates.
• The core loop was built inside a single Codex thread.

Inference The technical stack suggests a minimal viable prototype with strong attention to sandboxing and execution safety. However, the lack of operational data or deployment details indicates no production-grade delivery.

The description states:

• The project was submitted to the OpenAI 2026 hackathon.
• It was built in a short timeframe (a hackathon).
• The author notes that discard rates were lower than expected, which changed the product’s framing.
• No mention of users, customers, or adoption.

Not evidenced There is no evidence of traction, revenue, or user engagement beyond the author's own development and testing. No data on usage, retention, or impact is available.

The description states:

• The project addresses a gap in language model-generated coding exercises.
• It leverages Codex and GPT-5.6 for generation.
• It introduces sandboxing as a verification method.

Inference While the idea of verifying AI-generated content is not new, this approach uses sandboxed execution to validate code correctness. However, no comparison with existing tools or platforms in the coding education space is made.

The description states:

• The project was built as a hackathon submission.
• It relies on Codex credits rather than API access.
• There is no evidence of commercial viability or scalability.
• No data on real-world performance or adoption exists.

Inference Key risks include lack of funding, unclear monetization, and absence of real-world testing. The product may not be scalable beyond the author’s own use case.

1. What is your plan for scaling beyond a single-person hackathon prototype?
2. How do you intend to monetize this tool if it's only used by educators?
3. Have you tested the system with actual students or teachers?
4. Is there any evidence of interest from educational institutions or platforms?
5. Can you describe how the discard log is intended to be used in practice?