AI use disclosure

gasguide uses AI to scale production of educational content — narration, illustrations, voice mock interviews — but every clinical fact, drug dose, and algorithm is hand-curated and source-cited by a human. AI does the grunt work; CRNAs do the review. We tell you exactly where AI is used so you can judge it for yourself.

• Lecture narration audio. Lecture audio is generated with Kokoro TTS (open-source text-to-speech) from human-written scripts. The script is the source of truth; the voice is synthesized for production speed and consistency.
• Lecture keyframes + medical illustrations. Diagrams, illustrations, and visual aids are produced with FLUX (an open-source image-generation model). Each one is reviewed before publishing; visuals that fail anatomical accuracy or educational quality are rejected and re-rendered. Source images and prompts are auditable internally.
• Voice mock interviews + oral boards. The examiner persona is implemented with an open-source LLM (Qwen 2.5 32B locally hosted). Scoring rubrics are written by a CRNA. Whisper (open-source) handles your speech-to-text. Audio you submit is processed in real time and not used to train any model.
• Customer-service chatbot.The in-app chatbot answers product questions using retrieval over our public docs. It can't answer clinical questions or modify your account. Edge cases route to a human at support@gasguide.app.

• Q-bank questions, explanations, sources.Every one of the hand-curated questions in the bank is human-authored with explicit citation to Miller's, Stoelting, Vargo, AANA Practice Manual, MHAUS, ASRA, ACOG, AHA, or equivalent primary source. No batch AI fills.
• Drug doses, ranges, contraindications, reversal protocols. Manually cross-checked against package inserts and primary references. AI is not in the critical path.
• Crisis algorithms. Hand-authored from MHAUS, ASRA, ACOG, AHA, ATLS, and ASA practice advisories. No generative AI in the steps, drugs, or pitfalls.
• Calculator formulas. Hand-coded from primary literature.

• Your study attempts, credentials, mock-interview transcripts, and uploaded documents are not used to train any AI model. Period.
• We don't share your audio, transcripts, or written responses with third-party AI providers for training. The voice loop runs on infrastructure we control.
• Anonymized aggregate analytics (e.g., "25% of users miss Q-bank item bs-202") are used to improve content. Individual identifiers are not part of that analysis.

Everything on gasguide is educational. No content here — including AI-narrated lectures, AI-illustrated diagrams, voice mock interview feedback, or chatbot answers — substitutes for clinical judgment, your facility's protocols, primary sources, or supervision by a licensed anesthesia provider.

We make mistakes. If you find a wrong dose, a flawed algorithm step, an inaccurate illustration, or an AI-narrated word that's clinically wrong, email errata@gasguide.app with the page URL and we'll fix it within 7 days. Material errors in clinical content get a published changelog entry.