Ariad: Breast

The description states that Ariad: Breast is a patient-facing educational prototype for adults receiving systemic therapy for breast cancer. It offers two entry points:

• I’m starting treatment
• I’m having a symptom

For those starting treatment, users can search for drugs or regimens and review three practical questions: what to know before treatment, what may help at home, and when to contact the cancer team.

For those experiencing symptoms, users describe their symptom in everyday language, select from a controlled list of categories, add treatment context, and answer factual questions. The system then presents fixed, source-controlled education and warning information, along with a neutral summary of the facts entered, which can be copied or shared with the cancer team.

The tool uses GPT-5.6 in two tightly bounded roles:

1. Matching unfamiliar wording to symptom IDs from its catalogue.
2. Restating supplied facts into a neutral summary.

It does not diagnose symptoms, assign toxicity grades, recommend treatment changes, or write clinical guidance.

The interface is built with Next.js and React, deployed on Cloudflare Workers. It reads one compiled release at runtime, and all logic is handled by pure functions rather than generated on demand.

Not evidenced: The tool does not support accounts, data collection, or server-side symptom history.

The author states that Ariad: Breast grew from two ideas:

1. Making complex treatment information easier to navigate.
2. Building the limits into the product from the beginning.

It is positioned as a calm path through a difficult moment, aiming to help patients move from what they are noticing to understandable, source-linked information they can discuss with their cancer team.

The tool explicitly avoids:

• Improvising medical guidance
• Making plausible-sounding but incorrect answers
• Replacing the treating cancer team as the primary source of help

It is described as a clinical boundary tool, not a chatbot or triage system. The author emphasizes that it does not attempt to determine causation, assign urgency, or recommend changes.

The positioning reflects an intent to be transparent and safe, rather than comprehensive or predictive.

The target customer is defined as adults receiving systemic therapy for breast cancer.

The tool is described as a patient-facing educational prototype, not intended for clinicians or caregivers. It is built for individuals navigating treatment-related symptoms or preparing for treatment.

Not evidenced: No information on whether the tool targets specific subgroups (e.g., early-stage vs. metastatic), age ranges, or language preferences beyond English.

The description does not provide any evidence of a business model or pricing strategy.

It is described as a prototype, and no mention is made of monetization, licensing, or customer acquisition plans.

Not evidenced: No revenue streams, pricing tiers, or commercial partnerships are mentioned.

The product is built with:

• Next.js and React
• Codex as the sole engineering collaborator
• A versioned, source-controlled clinical content architecture
• Pure functions for logic, not dynamic generation
• Deterministic compiler that produces immutable releases with SHA-256 hashes

GPT-5.6 is used in bounded roles only, and all outputs are validated or fall back to deterministic behavior.

The system has:

• 175 unit and golden tests
• End-to-end patient journeys
• Adversarial model-output tests
• Accessibility checks
• Mobile testing down to 320 pixels

Deployment is on Cloudflare Workers with server-side API keys, rate limits, and closed fallbacks.

Not evidenced: No information on scalability, infrastructure costs, or long-term technical roadmap.

The project is described as a prototype, built during a hackathon (OpenAI 2026).

It includes:

• 492 exact object versions
• 139 patient-facing educational modules
• 101 source records
• 28 observable symptom concepts
• Three complete treatment-and-symptom demonstration pathways

The prototype supports preparation information for 53 drugs and eight regimens.

Not evidenced: No data on user adoption, engagement, or feedback. No evidence of real-world usage or clinical integration.

The description does not mention any direct competitors or market context beyond the general space of patient-facing health tools or symptom checkers.

It is described as a prototype, and no indication exists whether it competes with existing platforms or fills a gap in the market.

Not evidenced: No competitive analysis, market size, or positioning relative to other tools.

1. Unreviewed clinical content: The prototype is explicitly marked as unreviewed and draft. No clinical approval or review process is described beyond “next milestone.”
2. No commercialization plan: There is no evidence of a business model, pricing, or customer acquisition strategy.
3. Limited scope: The tool only supports 53 drugs and eight regimens, with three demonstration pathways. It does not appear to be scalable beyond its current scope.
4. No data collection or analytics: The system does not collect user data or store symptom histories, which may limit future development or clinical utility.
5. Dependency on GPT-5.6: While the model is used in bounded roles, any failure or unavailability could impact functionality.

1. What is the timeline and process for clinical review of the content?
2. Are there plans to expand beyond the current 53 drugs and eight regimens?
3. How will the tool be integrated into existing clinical workflows or systems?
4. Is there any plan for user feedback collection or iterative improvement?
5. What are the legal and regulatory considerations for deploying such a tool in a healthcare setting?
6. Are there any partnerships or collaborations with oncology institutions already in place?