User guide & employer evaluation
Job requirements
Design, refine and evaluate prompts for GPT, Claude and other LLMs; demonstrate zero-shot, few-shot, role prompting, CoT, ToT, ReAct, RAG, vector retrieval, agent workflows, Python, evaluation methodology, documentation and governance.
Workflow
- Administrator creates an account with an appropriate role.
- Candidate or Authenticator submits a prompt, model name and response, or uploads a supported text document.
- The local rubric checks prompt structure; it does not judge factual correctness.
- Authenticator reviews evidence and assigns a separate human score and critique.
- Search the archive and inspect metrics. Compare only equivalent test settings for performance claims.
- Use PEECAgent to explore the knowledge base and identify relevant records.
Employer critique checklist
Review authentication, prompt versioning, dataset coverage, grading validity, RAG evidence, prompt-injection resistance, privacy, reproducibility, documentation and upgrade readiness. Record strengths, limitations and recommended improvements.
Known limitations
Live GPT/Claude execution, semantic vector search, reliable automatic correctness judging, malware scanning, MFA, password recovery and production-grade deployment review are not included in this standalone foundation.
