import os
import re
from typing import Tuple, List
import gradio as gr

# =============================
# Config
# =============================
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "")
USE_OPENAI = bool(OPENAI_API_KEY)

# Prefer the most reliable model first, with fallbacks
PREFERRED_MODELS = ["gpt-4o", "gpt-4o-mini", "gpt-3.5-turbo"]

oai_client = None
if USE_OPENAI:
    try:
        from openai import OpenAI
        oai_client = OpenAI(api_key=OPENAI_API_KEY)
    except Exception:
        USE_OPENAI = False

# =============================
# Helpers
# =============================
def normalize_text(t: str) -> str:
    return re.sub(r"\s+", " ", (t or "").strip())

def keyword_match_score(resume: str, jd: str) -> Tuple[float, List[str]]:
    """Simple keyword overlap metric for demo."""
    def toks(s):
        return {w.lower() for w in re.findall(r"[A-Za-z][A-Za-z\-]{3,}", s)}
    r_set = toks(resume)
    j_set = toks(jd)
    if not r_set or not j_set:
        return 0.0, []
    overlap = sorted(list((r_set & j_set)))
    score = 100.0 * len(overlap) / max(1, len(j_set))
    return round(score, 1), overlap[:30]

def call_openai_with_fallback(system_prompt: str, user_prompt: str) -> str:
    """Try several models; return first successful response or empty string."""
    if not (USE_OPENAI and oai_client):
        return ""
    for model in PREFERRED_MODELS:
        try:
            resp = oai_client.chat.completions.create(
                model=model,
                messages=[
                    {"role": "system", "content": system_prompt},
                    {"role": "user", "content": user_prompt},
                ],
                temperature=0.5,
            )
            content = resp.choices[0].message.content
            if content:
                return content.strip()
        except Exception:
            # Try next model in the list
            continue
    return ""

# =============================
# Prompts
# =============================
SYS_BASE = (
    "You are CareerForge, a concise, supportive career assistant. "
    "Prioritize clarity, actionability, fairness, and transparency. "
    "Never fabricate facts about the candidate."
)

def prompt_resume_improvements(resume: str, jd: str) -> str:
    return (
        f"Job Description:\n{jd}\n\n"
        f"Resume:\n{resume}\n\n"
        "Task:\n"
        "1) Give a brief fit summary (2–3 sentences).\n"
        "2) List 6–8 bullet-point improvements grouped under headings: "
        "'Impact Metrics', 'Skills & Keywords', 'Experience Rewording', 'ATS Hygiene'.\n"
        "3) Provide an optimized 3–5 line professional summary."
    )

def prompt_cover_letter(resume: str, jd: str) -> str:
    return (
        f"Job Description:\n{jd}\n\n"
        f"Resume:\n{resume}\n\n"
        "Generate a 3-paragraph cover letter draft (150–220 words): "
        "(1) hook + role match, (2) 2–3 aligned achievements, (3) brief close."
    )

def prompt_mock_interview(resume: str, jd: str, style: str) -> str:
    return (
        f"Style: {style}. Ask 5 questions tailored to the job description and resume.\n\n"
        f"Job Description:\n{jd}\n\n"
        f"Resume:\n{resume}\n\n"
        "For each question, provide:\n"
        "- The QUESTION\n"
        "- A STRONG SAMPLE ANSWER (6–8 sentences)\n"
        "- A 1–2 sentence FEEDBACK TIP"
    )

BIAS_REFLECTION = (
    "### Bias & Fairness Reflection\n\n"
    "Risks:\n"
    "• Over-weighting prestigious schools or employers\n"
    "• Penalizing career gaps\n"
    "• Favoring culture-specific language\n"
    "• Keyword stuffing bias\n\n"
    "Mitigations:\n"
    "• Neutral, skill-first phrasing\n"
    "• Prompts discourage demographic inferences\n"
    "• Encourage transferable skills\n"
    "• Transparent keyword match score\n\n"
    "Limitations: Human review is always essential."
)

# =============================
# Core Functions
# =============================
def ingest_inputs(jd_text: str, resume_text: str):
    jd_clean = normalize_text(jd_text)
    resume_clean = normalize_text(resume_text)
    score, matched = keyword_match_score(resume_clean, jd_clean)
    visual = f"Match Score: {score}/100\nTop Overlaps: {', '.join(matched) if matched else '—'}"
    return jd_clean, resume_clean, visual

def generate_improvements(jd_clean: str, resume_clean: str):
    if not jd_clean or not resume_clean:
        return "Please provide both Job Description and Resume."
    out = call_openai_with_fallback(SYS_BASE, prompt_resume_improvements(resume_clean, jd_clean))
    if out:
        return out
    # Mock fallback
    score, matched = keyword_match_score(resume_clean, jd_clean)
    return (
        f"(Mock Output)\nResume aligns moderately (score {score}).\n\n"
        "Improvements:\n"
        "• Impact Metrics: add measurable outcomes (e.g., “reduced processing time by 18%”).\n"
        "• Skills & Keywords: surface real skills from the JD: " + ", ".join(matched[:8]) + "\n"
        "• Experience Rewording: lead with strong verbs; include tools.\n"
        "• ATS Hygiene: standard headers; avoid images/tables.\n\n"
        "Optimized Summary: Results-driven candidate aligning skills and impact with the posted role."
    )

def generate_cover_letter(jd_clean: str, resume_clean: str):
    if not jd_clean or not resume_clean:
        return "Please provide both Job Description and Resume."
    out = call_openai_with_fallback(SYS_BASE, prompt_cover_letter(resume_clean, jd_clean))
    if out:
        return out
    return (
        "(Mock Cover Letter)\n"
        "Dear Hiring Manager,\n\n"
        "I’m excited to apply for this role. My background in process improvement and collaboration aligns with your needs. "
        "Recently, I contributed to a workflow update that improved turnaround time and reporting clarity.\n\n"
        "I enjoy translating goals into actionable plans and using data to measure outcomes. "
        "I’m eager to support your team’s objectives and would welcome the chance to discuss how I can help.\n\n"
        "Sincerely,\nCandidate"
    )

def generate_mock_interview(jd_clean: str, resume_clean: str, style: str):
    if not jd_clean or not resume_clean:
        return "Please provide both Job Description and Resume."
    out = call_openai_with_fallback(SYS_BASE, prompt_mock_interview(resume_clean, jd_clean, style))
    if out:
        return out
    return (
        f"(Mock {style} Interview)\n\n"
        "Q1: Tell me about a project relevant to this role.\n"
        "A: I mapped requirements, built a quick prototype, iterated with feedback, and reduced cycle time.\n"
        "Tip: Add concrete metrics and tools.\n\n"
        "Q2: How do you prioritize conflicting tasks?\n"
        "A: Clarify goals, assess impact/urgency, time-box experiments, and communicate trade-offs early.\n"
        "Tip: Give a brief example with outcome.\n\n"
        "Q3: Describe a time you handled ambiguity.\n"
        "A: I proposed a small experiment, gathered data, and adjusted scope based on results.\n"
        "Tip: Share what you learned.\n\n"
        "Q4: How do you measure success?\n"
        "A: Define indicators upfront and review them with stakeholders.\n"
        "Tip: Tie metrics to the JD.\n\n"
        "Q5: Why this role/company?\n"
        "A: The mission, product focus, and my experience align with the team’s goals.\n"
        "Tip: Reference 1–2 specifics from the JD."
    )

# =============================
# Demo Defaults
# =============================
DEMO_JD = "Responsibilities: Collaborate, improve processes, analyze metrics. Preferred: Python, SQL."
DEMO_RESUME = "Experience: Assisted in process improvement; created reports. Skills: Python, SQL, Excel."

# =============================
# UI
# =============================
with gr.Blocks(title="CareerForge — Resume & Interview Coach") as demo:
    gr.Markdown("# CareerForge — Resume & Interview Coach")
    gr.Markdown(
        "Paste a **Job Description** and **Resume**, then generate suggestions, a cover letter, and a mock interview.\n"
        f"{'🔐 OpenAI mode enabled.' if USE_OPENAI else '⚠️ Mock Mode (no OPENAI_API_KEY set).'}"
    )

    with gr.Tab("1) Inputs"):
        with gr.Row():
            jd_text = gr.Textbox(label="Job Description (paste)", value=DEMO_JD, lines=10)
            resume_text = gr.Textbox(label="Resume (paste)", value=DEMO_RESUME, lines=10)
        process_btn = gr.Button("Ingest & Score")
        jd_out = gr.Textbox(label="Job Description (processed)", interactive=False)
        resume_out = gr.Textbox(label="Resume (processed)", interactive=False)
        match_vis = gr.Textbox(label="Keyword Match Score", interactive=False)

    with gr.Tab("2) Resume Improvements"):
        improve_btn = gr.Button("Generate Suggestions")
        improve_out = gr.Markdown()

    with gr.Tab("3) Cover Letter"):
        cover_btn = gr.Button("Generate Cover Letter")
        cover_out = gr.Markdown()

    with gr.Tab("4) Mock Interview"):
        style = gr.Dropdown(choices=["Behavioral", "Technical", "Product/Program"], value="Behavioral", label="Interview Style")
        interview_btn = gr.Button("Start Mock Interview")
        interview_out = gr.Markdown()

    with gr.Tab("5) Bias & Fairness"):
        gr.Markdown(BIAS_REFLECTION)

    # Wiring
    process_btn.click(fn=ingest_inputs, inputs=[jd_text, resume_text], outputs=[jd_out, resume_out, match_vis])
    improve_btn.click(fn=generate_improvements, inputs=[jd_out, resume_out], outputs=[improve_out])
    cover_btn.click(fn=generate_cover_letter, inputs=[jd_out, resume_out], outputs=[cover_out])
    interview_btn.click(fn=generate_mock_interview, inputs=[jd_out, resume_out, style], outputs=[interview_out])

# Local dev
if __name__ == "__main__":
    demo.launch()