How we use AI

DecipherU uses AI. Here's what, where, and what a human does before anything reaches you.

Authority for AI integration decisions

Decisions about AI integration on DecipherU are made by the founder, Julian Calvo, who is currently completing graduate-level training in applied AI specializing in cybersecurity at Northeastern University. AI use is bounded by the published editorial policy and methodology review standards. The Northeastern specialization track informs which AI capabilities ship into product features and which stay behind editorial review.

Content production

Our cybersecurity career guides, certification guides, salary pages, glossary entries, and research articles are written with AI assistance. Typical workflow: an editor defines the outline from primary sources (BLS, O*NET, NIST, certifying body pages, SEC filings), an AI model drafts sections, and the editor revises the draft for accuracy, voice, and compliance with our editorial policy.

No AI-generated draft is published without human review. Every page ships with a visible "Last verified" date set by the reviewing editor.

Models we use

• Claude by Anthropic. Primary content drafting, translation, summarization.
• GPT-4 / GPT-5 by OpenAI. Secondary drafting, structured-data generation.
• We do not use AI models that train on customer data without our consent. All API use is configured for zero data retention with the provider where available.

Product features powered by AI

• Range AI adversary system. The Range training environment uses the Anthropic API to drive in-game adversary decisions, dialogue, and incident escalation paths. Each scenario is bounded by a deterministic scenario script; the AI fills in adversary tactics within those bounds.
• AI Coach. The AI Coach in your dashboard uses the Anthropic API to hold career planning conversations grounded in your DecipherU profile, assessment history, and the public career guides. Coach responses are advisory and do not replace human career counseling.
• Embedding-based semantic search. Planned. When live, search across career guides, certifications, and assessments will use vector embeddings of the indexed content, plus a deterministic re-ranker. We will update this disclosure when search ships.
• AI Impact Outlook scoring. The AI Impact Outlook score uses a deterministic rubric, not a per-request AI call. The 6 dimensions are scored against a fixed model that we publish on the methodology page. Editorial review updates the rubric quarterly. Important: the score itself is not generated by an AI model at request time.
• AI Resume Review and AI Mock Interview tools return evaluations generated by language models with human-designed rubrics. Output is educational and does not guarantee interview results.
• Assessment scoring uses statistical models plus AI-generated narrative summaries of your profile. The underlying scores are deterministic; the summary text is generated and clearly labeled.
• Career Guide chat widgets answer questions grounded in the site's indexed content. When the model is uncertain, it says so and links to the authoritative page.

Anthropic data residency and training

When DecipherU sends user input (a Coach message, a Range scenario response, an assessment answer) to the Anthropic API, that input is processed by Anthropic to return our result. Per Anthropic's data-use policy for API customers, user inputs to the Anthropic API are not used to train Anthropic's models. Anthropic processes the request and returns a response; the data is not retained for training.

Inputs may be retained by Anthropic for a short operational window (for abuse monitoring), then deleted per their published policy. DecipherU does not export user data to any other AI provider.

Translations

Our Spanish (/es/*) and Brazilian Portuguese (/pt-br/*) landing pages and legal summaries are written natively, not machine-translated. Where we use AI-assisted translation for inner content pages, each translated page is flagged with a translation-quality score and reviewed before it replaces the English fallback. Pages scoring below our threshold stay as English until reviewed.

What AI does not decide

• Salary figures, certification costs, and company facts are copied from primary sources, not generated.
• Rankings and "best-for" judgments on comparison pages reflect editorial analysis, not AI opinion.
• Hiring or employment outcomes. Our tools support your decisions; they don't make them.

Your data

Your assessment responses, resume uploads, and chat messages are processed by our AI providers for the purpose of returning your result. We do not permit providers to train models on your data. See our privacy policy for complete detail on data handling.

Limitations and accuracy

Language models can hallucinate, repeat outdated facts, or invent citations. We treat every AI-drafted page as a starting point that requires founder review before publishing. Every cybersecurity content piece carries a sources_cited field, and a quarterly accuracy retrospective is published at /outcomes. If you find an inaccuracy, the line of accountability ends with the founder byline.

User feedback mechanism

Every cybersecurity content page carries a "Report an inaccuracy" link that opens a pre-filled message to editorial@decipheru.com. Confirmed errors are corrected within 48 hours of receipt and logged in the public changelog. A quarterly correction summary publishes alongside the outcomes report.

Questions

editorial@decipheru.com