/**
 * Pre-scripted AI tutor responses for the prototype.
 * Each pair has a Coach (motivational, pacing) and Tutor (concept, Socratic)
 * variant keyed by intent.
 */

export interface TutorResponse {
  id: string;
  intent: string;
  agent: 'coach' | 'tutor';
  message: string;
  suggestedActions?: string[];
}

export const aiTutorResponses: TutorResponse[] = [
  {
    id: 'resp-001',
    intent: 'stuck-on-concept',
    agent: 'coach',
    message:
      "I can see you've hit a wall on RAG Pipeline Design — that's the most common sticking point in this stack. Take a 5-minute break, then come back and let's try a smaller chunk. You've already mastered Tool Use, so the retrieval pieces will click faster than you expect.",
    suggestedActions: ['Take a break', 'Review a worked example', 'Break it into smaller steps'],
  },
  {
    id: 'resp-002',
    intent: 'stuck-on-concept',
    agent: 'tutor',
    message:
      "Let's build intuition first. RAG is a two-stage retrieval pipeline: (1) given a query, find the most relevant passages from a corpus, and (2) feed those passages into the model as context so its answer is grounded. Socratic question: why might a lexical search (like BM25) struggle with a query phrased very differently from the document text?",
    suggestedActions: ['Show me a worked example', 'Compare BM25 vs embeddings', 'Try a practice problem'],
  },
  {
    id: 'resp-003',
    intent: 'request-pacing',
    agent: 'coach',
    message:
      "You're 62% through the AI Orchestration stack and ahead of your original pace by nine days. Your next two defenses are scheduled in 11 days. I recommend a 3-day sprint on Multi-Agent Communication, then one light day before the defense prep.",
    suggestedActions: ['Show my weekly plan', 'Schedule a lighter day', 'Adjust pace'],
  },
  {
    id: 'resp-004',
    intent: 'request-pacing',
    agent: 'tutor',
    message:
      'Here is a suggested sequence: complete the Multi-Agent Communication module in 3 sessions of 45 minutes, then the Memory Systems primer in 2 sessions. After that, one full session for defense rehearsal. This matches the cognitive load profile of learners who succeeded before you.',
    suggestedActions: ['Lock in this plan', 'Adjust session length', 'See alternatives'],
  },
  {
    id: 'resp-005',
    intent: 'celebrate-milestone',
    agent: 'coach',
    message:
      "Huge milestone — you just earned your Tool Use & Function Calling microcredential with a 91. That's above the cohort median. You now have three of the five core orchestration credentials. Two more and you qualify for the stack endorsement.",
    suggestedActions: ['See path to endorsement', 'Share with employer partners', 'Pick the next competency'],
  },
  {
    id: 'resp-006',
    intent: 'celebrate-milestone',
    agent: 'tutor',
    message:
      'Your artifact for this competency showed clean handling of structured outputs and graceful fallback when the model returned malformed JSON. For your next competency, I recommend exploring how tool-use composes with multi-agent communication — that will set you up for the stack endorsement.',
    suggestedActions: ['Start Multi-Agent Communication', 'Review a harder artifact example', 'Read a case study'],
  },
  {
    id: 'resp-007',
    intent: 'explain-socratic',
    agent: 'tutor',
    message:
      "Think about agent memory as three layers: short-term context (the current conversation window), episodic memory (records of past agent runs you can replay), and long-term semantic memory (consolidated knowledge you distill over time). Question: if an agent handles 10,000 sessions a day, which layer would you consolidate into long-term memory, and how often?",
    suggestedActions: ['I have an answer', 'Give me a hint', 'Show a worked example'],
  },
  {
    id: 'resp-008',
    intent: 'defense-prep',
    agent: 'coach',
    message:
      "Your oral defense for Multi-Agent Communication is in 11 days. You are well prepared — your process trace shows 14 distinct practice runs. Let's spend one session rehearsing the edge-case questions, then a mock defense with me acting as the panel.",
    suggestedActions: ['Start mock defense', 'Review edge-case questions', 'See my process trace'],
  },
  {
    id: 'resp-009',
    intent: 'defense-prep',
    agent: 'tutor',
    message:
      "In a defense you'll be asked to explain your design trade-offs, not just your code. Be ready to answer: why did you choose a blackboard architecture over direct message passing? What failure mode did you observe under load, and how did you mitigate it? Let's rehearse one question now.",
    suggestedActions: ['Rehearse question 1', 'Rehearse question 2', 'See scoring rubric'],
  },
  {
    id: 'resp-010',
    intent: 'career-guidance',
    agent: 'coach',
    message:
      "Based on your competencies and artifacts, you match strongly to AI Orchestration Engineer roles (96%) and Agent Reliability Engineer roles (90%). Two employers are actively hiring for these profiles. Want me to show you the matching jobs and the remaining competencies they require?",
    suggestedActions: ['Show matching jobs', 'See competency gaps', 'Build a targeted plan'],
  },
  {
    id: 'resp-011',
    intent: 'career-guidance',
    agent: 'tutor',
    message:
      'Your portfolio demonstrates multi-agent systems and evaluation, which are the two most-cited skills in senior orchestration postings. The gap to a Staff-level role is observability and cost optimization. I recommend the Agent Reliability Engineer competencies as your next sprint.',
    suggestedActions: ['Start reliability sprint', 'See a staff-level job', 'Compare skill gaps'],
  },
  {
    id: 'resp-012',
    intent: 'check-understanding',
    agent: 'tutor',
    message:
      'Quick check: in a plan-and-execute agent, what is the advantage of re-planning after each tool call rather than executing the full plan from the start? Take your time — there is no penalty for thinking.',
    suggestedActions: ['I have an answer', 'Give me a hint', 'Skip this check'],
  },
];