import requests
import os
import gradio as gr
from datetime import datetime
import json
import uuid

APP_TITLE = "HumAI Midfielder Avatar"
APP_VERSION = "v0.2.0-enterprise-demo"
INFERENCE_URL = os.getenv("INFERENCE_URL", "")
INFERENCE_API_KEY = os.getenv("INFERENCE_API_KEY", "")
LIVE_PRODUCT_URL = "https://humai-orchestration-makerfire.vercel.app"
BRAND_LAYER = "BPM RED Academy / MightHub HumAI Layer"
PRODUCT_NAME = "HumAI Midfielder Avatar"
DOA_NAME = "MightHub DOA"
DOA_FULL_NAME = "MightHub DOA — Duty Officer Avatar"

DOMAINS = {
    "Urban Mobility": "urban",
    "Startup Scaling": "startup",
    "Public Systems": "public",
    "Finance / Compliance": "finance",
}

MODES = {
    "Decision Support": "decision",
    "Risk Assessment": "risk",
    "Optimization": "optimization",
    "AI Dispatcher": "dispatch",
    "Duty Officer Support": "duty",
}

SCENARIOS = {
    "Sarajevo congestion after work hours": "congestion",
    "Startup funding and resource allocation": "funding",
    "Public service coordination under pressure": "service",
    "Financial risk and compliance review": "compliance",
}

PRIORITIES = [
    "Speed",
    "Cost",
    "Comfort",
    "Sustainability",
    "Safety",
    "Urgency",
    "Risk control",
    "Investor readiness",
    "Transparency",
    "Accountability",
    "Operational continuity",
    "Human review",
]


def normalize_selection(value, mapping, fallback):
    return mapping.get(value, fallback)


def clamp(value, min_value=0.52, max_value=0.96):
    return max(min_value, min(max_value, value))

def call_real_inference(prompt):

    if not INFERENCE_URL:
        return {
            "success": False,
            "fallback": True,
            "content": "Inference endpoint not configured."
        }

    headers = {
        "Authorization": f"Bearer {INFERENCE_API_KEY}",
        "Content-Type": "application/json"
    }

    payload = {
        "model": "FinC2E",
        "messages": [
            {
                "role": "system",
                "content": "You are FinC2E governance runtime."
            },
            {
                "role": "user",
                "content": prompt
            }
        ],
        "temperature": 0.1,
        "max_tokens": 400
    }

    try:

        response = requests.post(
            INFERENCE_URL,
            headers=headers,
            json=payload,
            timeout=60
        )

        data = response.json()

        content = (
            data.get("choices", [{}])[0]
            .get("message", {})
            .get("content", "")
        )

        return {
            "success": True,
            "fallback": False,
            "content": content,
            "raw": data
        }

    except Exception as e:

        return {
            "success": False,
            "fallback": True,
            "content": str(e)
        }


def build_avatar_intro(domain_label, mode_label, scenario_label, priority):
    return (
        f"{PRODUCT_NAME} online.\n\n"
        f"**{DOA_FULL_NAME}** is active as the human-facing dispatcher layer of MightHub.\n\n"
        f"I understand that you selected **{domain_label}** with **{mode_label}** "
        f"for the scenario **{scenario_label}**.\n\n"
        f"Your stated priority is **{priority}**. "
        "I will receive the intent, read the operational situation, route it through "
        "Mission Control, and return explainable decision support."
    )


def evaluate_mission_control(domain, mode, scenario, priority, user_context):
    risk = "MEDIUM"
    score = 0.72
    real_runtime = call_real_inference(user_context)
    recommendation = "Use structured Human-AI orchestration before taking operational action."
    explanation = (
        "HumAI structures the situation, evaluates domain context and prepares "
        "a transparent recommendation for human review."
    )
    impact = "Improves clarity, reduces decision friction and creates an auditable decision trail."
    operator_note = "Human review remains required before real-world operational execution."
    next_best_action = (
        "Capture the context, review the recommendation and decide whether additional data "
        "or human escalation is needed."
    )
    why_this_matters = (
        "Unstructured decisions create confusion. MightHub turns fragmented context into "
        "a readable operational picture."
    )
    avatar_response = (
        "I can support this as the Duty Officer Avatar by asking clarifying questions, "
        "structuring the decision and translating the output into human-readable guidance."
    )
    demo_pitch_line = (
        "This shows how HumAI Midfielder Avatar routes human intent into MightHub "
        "Mission Control instead of simply generating chatbot-style answers."
    )
    duty_officer_assessment = (
        "Initial watch-floor assessment: the situation is suitable for structured decision "
        "support with human review before action."
    )

    if domain == "urban":
        recommendation = (
            "Recommend the most realistic mobility option by balancing travel time, "
            "congestion pressure, user priority, cost and sustainability."
        )
        explanation = (
            "The system treats Sarajevo congestion as a mobility decision problem, not as "
            "a simple navigation question. It structures user context, priority and constraints "
            "before recommending action."
        )
        impact = "Supports smarter urban movement, reduced congestion pressure and clearer citizen guidance."
        operator_note = "Best demonstrated as a Sarajevo AI mobility dispatcher scenario."
        next_best_action = (
            "Ask the user whether speed, cost, comfort, safety or sustainability is the highest "
            "priority, then route the recommendation accordingly."
        )
        why_this_matters = (
            "Urban mobility decisions are usually made under pressure. A structured AI dispatcher "
            "can reduce uncertainty and help people choose better options in real time."
        )
        avatar_response = (
            "I will act as a Sarajevo mobility Duty Officer Avatar. Before recommending a route "
            "or option, I need to understand whether your priority is speed, cost, comfort, safety "
            "or sustainability."
        )
        duty_officer_assessment = (
            "Mobility watch assessment: the user requires route-oriented guidance, but the "
            "recommendation should remain explainable and priority-aware."
        )
        demo_pitch_line = (
            "In this scenario, HumAI Midfielder Avatar becomes a Sarajevo mobility dispatcher: "
            "it receives user context, reads the mobility field and routes an explainable "
            "recommendation through MightHub Mission Control."
        )

    if domain == "startup":
        recommendation = (
            "Prioritize resource allocation by separating urgent survival needs from strategic "
            "growth activities and investor-readiness work."
        )
        explanation = (
            "The system structures startup uncertainty into runway, traction, operational focus "
            "and investor narrative."
        )
        impact = (
            "Improves founder focus, reduces waste, strengthens fundraising preparation and helps "
            "the team communicate its operating logic."
        )
        operator_note = "Best used to show how HumAI supports founders, accelerators, mentors and early-stage investors."
        next_best_action = (
            "Identify current runway, strongest traction signal and highest-risk assumption before "
            "committing resources."
        )
        why_this_matters = (
            "Startups often fail because limited capital is spent without a clear operating logic. "
            "HumAI helps founders structure trade-offs before acting."
        )
        avatar_response = (
            "I will help structure this as a founder decision. We should clarify runway, traction, "
            "burn rate, investor readiness and the highest-risk assumption before taking action."
        )
        duty_officer_assessment = (
            "Startup watch assessment: the critical question is whether the team should protect "
            "runway, accelerate traction or prepare investor-facing evidence."
        )
        demo_pitch_line = (
            "In this scenario, HumAI Midfielder Avatar helps a startup move from uncertainty to "
            "an investor-ready decision narrative."
        )

    if domain == "public":
        recommendation = (
            "Structure the operational picture, classify incoming requests, identify bottlenecks "
            "and route decisions to the responsible human operator."
        )
        explanation = "The system supports public-service coordination without claiming autonomous authority."
        impact = "Improves transparency, accountability, response coordination and communication quality."
        operator_note = "Best used to demonstrate accountable public-system coordination."
        next_best_action = (
            "Classify requests by urgency, public impact and responsible unit, then escalate only "
            "the cases requiring human authority."
        )
        why_this_matters = (
            "Public systems need clarity, traceability and accountability. HumAI can support decision "
            "preparation while keeping people responsible."
        )
        avatar_response = (
            "I will structure this as a public-system coordination case. The goal is to clarify "
            "urgency, responsible unit, public impact and review boundary."
        )
        duty_officer_assessment = (
            "Public systems watch assessment: preserve accountability, classify requests and route "
            "only authority-dependent cases to human decision-makers."
        )
        demo_pitch_line = (
            "In this scenario, HumAI Midfielder Avatar acts as a Duty Officer layer for public service "
            "coordination, not as an autonomous authority."
        )

    if domain == "finance":
        recommendation = (
            "Classify risk, explain key indicators, preserve auditability and route the case toward "
            "human compliance review."
        )
        explanation = "The system structures compliance reasoning into risk, explanation, traceability and human review."
        impact = (
            "Supports structured compliance analysis, risk visibility, decision traceability and safer "
            "handling of sensitive financial or procurement cases."
        )
        operator_note = "Best used to explain FinC2E-style governance logic as a future specialized module."
        next_best_action = (
            "Separate factual indicators from assumptions, assign preliminary risk level and require "
            "human review before final disposition."
        )
        why_this_matters = (
            "Financial and compliance decisions require explainability. HumAI can help structure the "
            "case without replacing legal, financial or institutional authority."
        )
        avatar_response = (
            "I will structure this as a governance and compliance review. The goal is not autonomous "
            "enforcement, but explainable risk classification and human review."
        )
        duty_officer_assessment = (
            "Compliance watch assessment: maintain advisory-only boundaries, preserve auditability "
            "and route final disposition to human review."
        )
        demo_pitch_line = (
            "In this scenario, HumAI Midfielder Avatar demonstrates how compliance reasoning can be "
            "structured, explainable and human-reviewed."
        )

    if mode == "risk":
        score += 0.08
        recommendation += " Risk controls and documented human review should be applied."
        next_best_action = "Document the main risk drivers, identify missing information and route the case for responsible review."

    if mode == "optimization":
        score -= 0.06
        recommendation += " Optimization should focus on time, cost, operational load and measurable impact."
        next_best_action = "Compare the current process with the recommended action and remove the highest-friction step first."

    if mode == "dispatch":
        score += 0.03
        recommendation += " The recommendation should be delivered through a conversational dispatcher interface."
        next_best_action = "Convert the recommendation into a short, user-facing message that a conversational avatar or dispatcher can deliver clearly."

    if mode == "duty":
        score += 0.05
        recommendation += (
            " The Duty Officer Avatar should maintain the operational picture, ask clarifying questions "
            "and route the case to the correct decision boundary."
        )
        next_best_action = (
            "Summarize the situation, identify the missing field information, and route the case to "
            "the responsible human decision point."
        )

    if scenario == "congestion":
        risk = "MEDIUM"
        score += 0.04
    if scenario == "funding":
        risk = "HIGH"
        score += 0.09
    if scenario == "service":
        risk = "MEDIUM"
        score += 0.02
    if scenario == "compliance":
        risk = "HIGH"
        score += 0.10

    priority_lower = priority.lower()
    if priority_lower in ["urgency", "risk control", "accountability", "operational continuity", "human review"]:
        score += 0.03
    if priority_lower in ["sustainability", "transparency", "accountability"]:
        impact += " The selected priority also strengthens transparent, responsible and socially useful decision support."

    if user_context and len(user_context.strip()) > 0:
        explanation += " The user-provided context was considered as an additional narrative signal for the dispatcher response."
        avatar_response += f"\n\nBased on your note, I would first clarify: '{user_context.strip()[:180]}'"

    score = clamp(score)
    if score >= 0.80:
        risk = "HIGH"
    if score >= 0.90:
        risk = "CRITICAL"

    return {
        "session_id": str(uuid.uuid4()),
        "timestamp": datetime.utcnow().isoformat() + "Z",
        "engine": PRODUCT_NAME,
        "doa_layer": DOA_FULL_NAME,
        "version": APP_VERSION,
        "execution_mode": "deterministic_enterprise_fallback",
        "ai_assisted": real_runtime["success"],
        "inference_fallback": real_runtime["fallback"],
        "runtime_output": real_runtime["content"],
        "domain": domain,
        "mode": mode,
        "scenario": scenario,
        "priority": priority,
        "risk": risk,
        "confidence": round(score, 2),
        "human_review_required": True,
        "recommendation": recommendation,
        "explanation": explanation,
        "impact": impact,
        "operator_note": operator_note,
        "duty_officer_assessment": duty_officer_assessment,
        "next_best_action": next_best_action,
        "why_this_matters": why_this_matters,
        "avatar_response": avatar_response,
        "demo_pitch_line": demo_pitch_line,
        "product_boundary": (
            "This is a public MightHub DOA laboratory and deterministic demonstrator. "
            "It is not a certified production mobility, compliance or public-authority system."
        ),
    }


def render_markdown_output(decision):
    confidence = int(decision["confidence"] * 100)
    return f"""
# MightHub Mission Control Output

**Engine:** `{decision["engine"]}`  
**DOA Layer:** `{decision["doa_layer"]}`  
**Execution Mode:** `{decision["execution_mode"]}`  
**Domain:** `{decision["domain"]}`  
**Use Case:** `{decision["mode"]}`  
**Scenario:** `{decision["scenario"]}`  
**Priority:** `{decision["priority"]}`  

---

## Risk & Confidence

**Risk Level:** `{decision["risk"]}`  
**Confidence:** `{confidence}%`  
**Human Review Required:** `YES`

---

## Duty Officer Assessment

{decision["duty_officer_assessment"]}

---

## Recommended Action

{decision["recommendation"]}

---

## Explanation

{decision["explanation"]}

---

## Avatar Dispatcher Response

{decision["avatar_response"]}

---

## Next Best Action

{decision["next_best_action"]}

---

## Why This Matters

{decision["why_this_matters"]}

---

## Impact

{decision["impact"]}

---

## Operator Note

{decision["operator_note"]}

---

## Demo Pitch Line

> {decision["demo_pitch_line"]}

---

## Product Boundary

{decision["product_boundary"]}
"""


def run_humai_avatar(domain_label, mode_label, scenario_label, priority, user_context):
    domain = normalize_selection(domain_label, DOMAINS, "urban")
    mode = normalize_selection(mode_label, MODES, "dispatch")
    scenario = normalize_selection(scenario_label, SCENARIOS, "congestion")

    avatar_intro = build_avatar_intro(domain_label, mode_label, scenario_label, priority)
    decision = evaluate_mission_control(domain, mode, scenario, priority, user_context)

    avatar_panel = f"""
## {PRODUCT_NAME}

**Status:** Online  
**Role:** {DOA_FULL_NAME}  
**Current priority:** {priority}

{avatar_intro}

---

### Midfielder Doctrine

**Receive the intent. Read the field. Route the context. Support the human decision.**

---

### Next Question

**What matters most right now — speed, cost, comfort, safety, sustainability, urgency, operational continuity, transparency or risk control?**

The answer changes how MightHub Mission Control should prioritize the recommendation.
"""
    return avatar_panel, render_markdown_output(decision), json.dumps(decision, indent=2, ensure_ascii=False)


def clear_inputs():
    return (
        "Urban Mobility",
        "AI Dispatcher",
        "Sarajevo congestion after work hours",
        "Speed",
        "",
        f"## {PRODUCT_NAME}\n\n**Status:** Waiting for input\n\n**Role:** {DOA_FULL_NAME}\n\nSelect a domain, use case and scenario, then run the dispatcher.",
        "Mission Control output will appear here.",
        "{}",
    )


CUSTOM_CSS = """
.gradio-container {
    background: radial-gradient(circle at top left, rgba(34, 211, 238, 0.16), transparent 28%),
                radial-gradient(circle at bottom right, rgba(139, 92, 246, 0.16), transparent 30%),
                #020617 !important;
    color: #e2e8f0 !important;
}
#humai-hero {
    border: 1px solid rgba(34, 211, 238, 0.28);
    border-radius: 28px;
    padding: 28px;
    background: linear-gradient(135deg, rgba(8, 47, 73, 0.78), rgba(15, 23, 42, 0.94));
    box-shadow: 0 22px 80px rgba(8, 145, 178, 0.16);
}
#humai-hero h1 { font-size: 42px; line-height: 1.05; margin-bottom: 12px; }
#humai-hero p { color: #cbd5e1; font-size: 16px; line-height: 1.7; }
#signal-card {
    border: 1px solid rgba(34, 211, 238, 0.24);
    border-radius: 24px;
    padding: 20px;
    background: rgba(15, 23, 42, 0.82);
}
#signal-dot {
    width: 74px;
    height: 74px;
    border-radius: 999px;
    background: radial-gradient(circle, #67e8f9 0%, #0891b2 45%, rgba(8, 47, 73, 0.4) 100%);
    box-shadow: 0 0 38px rgba(103, 232, 249, 0.72);
    margin-bottom: 14px;
}
#doa-badge {
    display: inline-block;
    padding: 8px 12px;
    border-radius: 999px;
    border: 1px solid rgba(167, 139, 250, 0.45);
    background: rgba(91, 33, 182, 0.24);
    color: #ddd6fe;
    font-weight: 800;
    letter-spacing: 0.14em;
    text-transform: uppercase;
    font-size: 12px;
    margin-bottom: 12px;
}
textarea, input, select { border-radius: 16px !important; }
button { border-radius: 16px !important; font-weight: 800 !important; }
#footer-note { color: #94a3b8; font-size: 13px; line-height: 1.7; }
"""


with gr.Blocks(
    title=APP_TITLE,
    css=CUSTOM_CSS,
    theme=gr.themes.Soft(primary_hue="cyan", secondary_hue="violet", neutral_hue="slate"),
) as demo:
    gr.HTML(
        f"""
        <div id="humai-hero">
            <p style="letter-spacing: 0.28em; text-transform: uppercase; color: #67e8f9; font-weight: 800;">{BRAND_LAYER}</p>
            <div id="doa-badge">{DOA_NAME} / Duty Officer Avatar</div>
            <h1>{PRODUCT_NAME}</h1>
            <p>MightHub Human-AI dispatcher layer for Mission Control decisions, urban mobility intelligence, startup support, public-system coordination and governance-native AI workflows.</p>
            <p><strong>Operating doctrine:</strong> Receive the intent. Read the field. Route the context. Support the human decision.</p>
            <p>Live product interface: <a href="{LIVE_PRODUCT_URL}" target="_blank" style="color:#67e8f9; font-weight:800;">{LIVE_PRODUCT_URL}</a></p>
        </div>
        """
    )

    with gr.Row():
        with gr.Column(scale=1):
            gr.HTML(
                f"""
                <div id="signal-card">
                    <div id="signal-dot"></div>
                    <p style="letter-spacing:0.24em; text-transform:uppercase; color:#67e8f9; font-weight:800;">HumAI Midfielder Online</p>
                    <h2 style="margin-top:8px;">Receive. Read. Route.</h2>
                    <p style="color:#cbd5e1; line-height:1.7;">The Midfielder Avatar is the human interface to MightHub Mission Control. It receives intent, reads the situation, routes context into structured decisions and explains the output back to the user.</p>
                    <p style="color:#ddd6fe; line-height:1.7; font-weight:700;">{DOA_FULL_NAME}</p>
                </div>
                """
            )
            domain_input = gr.Dropdown(choices=list(DOMAINS.keys()), value="Urban Mobility", label="Domain")
            mode_input = gr.Dropdown(choices=list(MODES.keys()), value="AI Dispatcher", label="Use Case")
            scenario_input = gr.Dropdown(choices=list(SCENARIOS.keys()), value="Sarajevo congestion after work hours", label="Scenario")
            priority_input = gr.Dropdown(choices=PRIORITIES, value="Speed", label="Primary Priority")
            user_context_input = gr.Textbox(
                label="Optional User Context",
                placeholder="Example: I am near Marijin Dvor, I need to reach Ilidža, traffic is heavy and I care about cost and time.",
                lines=5,
            )
            with gr.Row():
                run_button = gr.Button("Run MightHub DOA", variant="primary")
                clear_button = gr.Button("Reset")

        with gr.Column(scale=1):
            avatar_output = gr.Markdown(
                label=PRODUCT_NAME,
                value=(f"## {PRODUCT_NAME}\n\n**Status:** Waiting for input\n\n**Role:** {DOA_FULL_NAME}\n\nSelect a domain, use case and scenario, then run the dispatcher."),
            )
            mission_output = gr.Markdown(label="MightHub Mission Control Output", value="Mission Control output will appear here.")

    with gr.Accordion("Structured JSON Output", open=False):
        json_output = gr.Code(label="Mission Control JSON", language="json", value="{}")

    gr.HTML(
        """
        <div id="humai-hero" style="margin-top: 24px;">
            <p style="letter-spacing: 0.28em; text-transform: uppercase; color: #a78bfa; font-weight: 800;">
                MightHub DOA System Card
            </p>

            <h2 style="font-size: 32px; line-height: 1.12; margin-bottom: 16px;">
                Duty Officer Avatar — Operational Boundary
            </h2>

            <p>
                MightHub DOA is designed as a human-facing dispatcher layer.
                It receives intent, reads the operational field, routes context
                into Mission Control and returns explainable decision support.
            </p>

            <div style="display: grid; grid-template-columns: repeat(auto-fit, minmax(240px, 1fr)); gap: 16px; margin-top: 22px;">
                <div id="signal-card">
                    <p style="letter-spacing:0.18em; text-transform:uppercase; color:#67e8f9; font-weight:800;">
                        What it does
                    </p>
                    <ul style="color:#cbd5e1; line-height:1.8;">
                        <li>Receives user intent</li>
                        <li>Reads domain and scenario context</li>
                        <li>Routes input into Mission Control</li>
                        <li>Produces structured recommendations</li>
                        <li>Explains next best action</li>
                    </ul>
                </div>

                <div id="signal-card">
                    <p style="letter-spacing:0.18em; text-transform:uppercase; color:#fbbf24; font-weight:800;">
                        What it does not do
                    </p>
                    <ul style="color:#cbd5e1; line-height:1.8;">
                        <li>Does not replace human judgment</li>
                        <li>Does not act autonomously</li>
                        <li>Does not issue public-authority decisions</li>
                        <li>Does not perform certified compliance review</li>
                        <li>Does not command real-world operations</li>
                    </ul>
                </div>

                <div id="signal-card">
                    <p style="letter-spacing:0.18em; text-transform:uppercase; color:#34d399; font-weight:800;">
                        Current mode
                    </p>
                    <ul style="color:#cbd5e1; line-height:1.8;">
                        <li>Deterministic enterprise fallback</li>
                        <li>Demo-safe behavior</li>
                        <li>Structured JSON output</li>
                        <li>Human review required</li>
                        <li>Public demonstrator boundary</li>
                    </ul>
                </div>

                <div id="signal-card">
                    <p style="letter-spacing:0.18em; text-transform:uppercase; color:#c084fc; font-weight:800;">
                        Future integrations
                    </p>
                    <ul style="color:#cbd5e1; line-height:1.8;">
                        <li>Gemini / Vertex AI reasoning</li>
                        <li>Firebase session memory</li>
                        <li>Google Maps mobility context</li>
                        <li>Azure AI Foundry evaluation</li>
                        <li>PitchAvatar or custom avatar layer</li>
                    </ul>
                </div>
            </div>
        </div>
        """
    )
gr.HTML(
        """
        <div id="footer-note">
            <p>
                <strong>Public demonstrator note:</strong>
                This Hugging Face Space currently runs as a deterministic enterprise
                MightHub DOA laboratory. It is designed for demo safety, explainability
                and future integration with Gemini / Vertex AI, Firebase, Google Maps,
                Azure AI Foundry, Hugging Face model artifacts, NVIDIA-oriented
                inference infrastructure and PitchAvatar or custom avatar layers.
            </p>

            <p>
                <strong>Product boundary:</strong>
                HumAI is advisory, human-in-the-loop and demonstration-oriented in this version.
                It does not replace human judgment, public authority, legal review or
                operational command.
            </p>
        </div>
        """
)


if __name__ == "__main__":
    demo.launch()