AI Disclosure & Ethics
Responsible AI for Defense Preparation
Transparent disclosure of how AI models generate questions, score answers, and protect your academic work.
How AI Panelists Work
PanelCrusher.AI uses large language models hosted via Groq and Hugging Face Inference endpoints to simulate academic defense panelists with distinct review personas (such as Friendly Advisor, Methodological Critic, Statistical Analyst, Technical Expert, and Strict Professor). Questions are grounded in the text extracted from your uploaded PDF.
Calibrated Scoring & Constructive Deliberation
Our evaluation engine scores answers on a 0–100 academic defense rubric where 75/100 is the passing benchmark. Coherent, relevant answers that address the panelist’s question and connect to your study’s methodology, findings, or scope earn passing marks (75–98), while constructive feedback highlights areas you can sharpen before your real defense.
Prompt Injection & Safety Safeguards
All document excerpts and user answers are isolated inside strict XML delimiters and evaluated against OWASP LLM01 prompt-injection safeguards. Attempts to override rubric instructions or inject system commands are automatically neutralized.
Human Oversight & Academic Integrity
AI-generated feedback is an educational rehearsal aid and may occasionally miss domain nuances. Always verify technical, statistical, ethical, and methodological recommendations with your thesis adviser, research committee, and institutional guidelines. PanelCrusher.AI is built to help you practice explaining your own research—not to fabricate research data or write thesis chapters for you.