app.py

"""
🔬 ForensiX AI — Digital Stratigraphy & Forensic Intelligence Platform
Multi-Page Dashboard with all winning features.
"""
import gradio as gr
import plotly.graph_objects as go
import plotly.express as px
from plotly.subplots import make_subplots
import pandas as pd
import numpy as np
import json, io
from datetime import datetime
from engine import ForensicOrchestrator, DualModeTODEstimator
import networkx as nx

orchestrator = ForensicOrchestrator()

DEMO_REPORT = """AUTOPSY REPORT - CASE #2024-MH-0847
DECEDENT: John Doe (unidentified male), Approximately 35-45 years
DATE OF EXAMINATION: March 15, 2024, 09:00 AM
DATE BODY FOUND: March 14, 2024, 06:30 PM
LOCATION: Abandoned warehouse, Industrial District, Block 7

EXTERNAL: Well-nourished adult male, ~78 kg. Rigor mortis fully developed. Lividity fixed posterior. Body temp (rectal): 28.5°C. Ambient: 18°C.

INJURIES:
1. Blunt force trauma to right temporal region, 4.5x3.2cm, with subdural hematoma
2. Defensive wounds on both forearms - three linear abrasions
3. Contusion on left shoulder, 6x4cm
4. Petechial hemorrhages in conjunctivae bilaterally
5. Ligature mark around neck, 0.5cm width

INTERNAL: Subdural hematoma (right hemisphere, ~80ml). Cerebral edema. Hyoid bone intact. Lungs mild congestion. Heart 340g. Stomach: partially digested meal.

TOXICOLOGY: Blood alcohol 0.04 g/dL. Benzodiazepines detected (trace - diazepam). No illicit substances detected.

CAUSE OF DEATH: Blunt force head trauma with subdural hematoma and asphyxia due to ligature compression of neck.
MANNER OF DEATH: Homicide

NOTES: Skin under fingernails collected for DNA analysis. Foreign fibers recovered (synthetic, blue). TOD estimated 12-18 hours prior to discovery."""

DEMO_EVIDENCE = """timestamp,source,event_type,location_lat,location_lon,details
2024-03-14T00:30:00,CCTV-Cam7,vehicle_detected,19.076,72.878,White sedan entering industrial area
2024-03-14T01:15:00,Mobile-Tower,cell_ping,19.0755,72.878,Victim phone connected tower ID-4421
2024-03-14T01:45:00,CCTV-Cam12,person_detected,19.0762,72.8775,Two individuals walking toward warehouse
2024-03-14T02:00:00,Mobile-Tower,cell_ping,19.0762,72.8774,Victim phone last active ping
2024-03-14T02:15:00,CCTV-Cam12,person_detected,19.0762,72.8775,Single individual leaving warehouse rapidly
2024-03-14T02:20:00,CCTV-Cam7,vehicle_detected,19.076,72.878,White sedan exiting at high speed
2024-03-14T02:30:00,Mobile-Tower,cell_disconnect,19.0762,72.8774,Victim phone disconnected
2024-03-14T06:00:00,CCTV-Cam7,person_detected,19.0758,72.878,Security guard patrol
2024-03-14T18:30:00,Emergency,call_received,19.0762,72.8775,Body discovered by security guard"""

def run_analysis(report, evidence_csv, state):
    global orchestrator
    orchestrator = ForensicOrchestrator()
    if not report.strip(): return {}, None, "⚠️ Provide report text.", state, None, None, ""
    evidence_list = []
    if evidence_csv.strip():
        df = pd.read_csv(io.StringIO(evidence_csv))
        evidence_list = df.to_dict('records')
    orchestrator.ingest_report(report)
    if evidence_list: orchestrator.ingest_evidence(evidence_list)
    results = orchestrator.run_analysis(report, evidence_list)
    risk_fig = build_risk_fig(results)
    graph_fig = build_graph_fig(results)
    timeline_fig = build_timeline_fig(results)
    summary = build_summary(results)
    state = {"results": results}
    chain_info = f"🔐 Chain: {'✅ VALID' if results['chain']['valid'] else '❌ BROKEN'} | {results['chain']['blocks']} blocks"
    return results, risk_fig, summary, state, graph_fig, timeline_fig, chain_info

def calc_tod(t_rec, t_amb, wt, corr, rigor, lividity, decomp, vk):
    est = DualModeTODEstimator()
    r = est.estimate(t_rec, t_amb, wt, corr, rigor, lividity, decomp, vk if vk > 0 else None)
    fig = go.Figure()
    if r.get("cooling_curve"):
        fig.add_trace(go.Scatter(x=[p["time"] for p in r["cooling_curve"]], y=[p["temp"] for p in r["cooling_curve"]],
            mode='lines', name='Cooling', line=dict(color='#60a5fa', width=3)))
        if r.get("pmi_hours"):
            fig.add_trace(go.Scatter(x=[r["pmi_hours"]], y=[t_rec], mode='markers', name='Measured',
                marker=dict(size=14, color='#ef4444', symbol='x')))
            fig.add_vrect(x0=max(0,r["pmi_hours"]-2.8), x1=r["pmi_hours"]+2.8, fillcolor="rgba(239,68,68,0.1)", line_width=0)
        fig.add_hline(y=t_amb, line_dash="dash", line_color="#22c55e")
    fig.update_layout(template="plotly_dark", paper_bgcolor="#0a0a0f", plot_bgcolor="#12121a",
        title=f"{'Henssge Model' if r.get('phase')=='early' else 'Late-Phase'}", height=380, font=dict(color="#e2e8f0"))
    md = f"## ⏱️ PMI: **{r.get('pmi_hours','N/A')} hours** ({r.get('phase','').title()} Phase)\n"
    md += f"**Method:** {r.get('method','')}\n**Confidence:** {r.get('confidence','')}\n"
    if r.get("lower_bound"): md += f"**95% CI:** {r['lower_bound']}–{r['upper_bound']}h"
    return r, fig, md

def handle_query(query, state):
    if not query.strip(): return "Enter a question."
    return orchestrator.query(query)

def load_demo(): return DEMO_REPORT, DEMO_EVIDENCE

def build_risk_fig(results):
    rf = [f for a in results.get("agents",[]) for f in a.get("findings",[]) if f.get("category")=="RISK_SCORE"]
    if not rf: return go.Figure()
    factors = rf[0].get("factors",{})
    cats = [k.replace("_"," ").title() for k in factors]; vals = list(factors.values())
    fig = make_subplots(rows=1, cols=2, specs=[[{"type":"polar"},{"type":"xy"}]], subplot_titles=("Risk Radar","Scores"))
    fig.add_trace(go.Scatterpolar(r=vals+[vals[0]], theta=cats+[cats[0]], fill='toself',
        fillcolor='rgba(239,68,68,0.2)', line=dict(color='#ef4444',width=2)), row=1, col=1)
    colors = ['#ef4444' if v>=70 else '#f59e0b' if v>=40 else '#22c55e' for v in vals]
    fig.add_trace(go.Bar(x=vals, y=cats, orientation='h', marker_color=colors, text=[f'{v:.0f}' for v in vals], textposition='inside'), row=1, col=2)
    fig.update_layout(template="plotly_dark", paper_bgcolor="#0a0a0f", plot_bgcolor="#12121a", height=380,
        showlegend=False, font=dict(color="#e2e8f0"), polar=dict(radialaxis=dict(range=[0,100], gridcolor="#1e293b"), bgcolor="#12121a"))
    fig.update_xaxes(range=[0,100], gridcolor="#1e293b", row=1, col=2)
    return fig

def build_graph_fig(results):
    gd = results.get("graph",{}); nodes = gd.get("nodes",[]); edges = gd.get("edges",[])
    if not nodes: return go.Figure()
    G = nx.DiGraph()
    for n in nodes: G.add_node(n["id"])
    for e in edges: G.add_edge(e["source"], e["target"])
    pos = nx.spring_layout(G, seed=42)
    ex, ey = [], []
    for e in edges:
        if e["source"] in pos and e["target"] in pos:
            ex.extend([pos[e["source"]][0], pos[e["target"]][0], None])
            ey.extend([pos[e["source"]][1], pos[e["target"]][1], None])
    colors_map = {"autopsy":"#ef4444","cctv":"#3b82f6","mobile":"#a855f7","iot":"#22c55e"}
    nx_list = [n for n in nodes if n["id"] in pos]
    fig = go.Figure()
    fig.add_trace(go.Scatter(x=ex, y=ey, mode='lines', line=dict(width=1,color='#475569'), hoverinfo='none'))
    fig.add_trace(go.Scatter(x=[pos[n["id"]][0] for n in nx_list], y=[pos[n["id"]][1] for n in nx_list],
        mode='markers+text', text=[f'{n["id"][:6]}' for n in nx_list], textposition='top center',
        textfont=dict(size=8,color='#94a3b8'), marker=dict(size=14, color=[colors_map.get(n["type"],"#64748b") for n in nx_list])))
    fig.update_layout(template="plotly_dark", paper_bgcolor="#0a0a0f", plot_bgcolor="#12121a", title="🔗 Evidence Graph",
        height=350, showlegend=False, xaxis=dict(showgrid=False,showticklabels=False), yaxis=dict(showgrid=False,showticklabels=False), font=dict(color="#e2e8f0"))
    return fig

def build_timeline_fig(results):
    tl = results.get("timeline",[])
    if not tl: return go.Figure()
    colors = {"autopsy":"#ef4444","cctv":"#3b82f6","mobile":"#a855f7","iot":"#22c55e"}
    fig = go.Figure()
    for stratum in set(e.get("stratum","") for e in tl):
        evs = [e for e in tl if e.get("stratum") == stratum and e.get("timestamp")]
        if evs:
            fig.add_trace(go.Scatter(x=[e["timestamp"] for e in evs], y=[stratum]*len(evs),
                mode='markers+text', name=stratum.upper(), text=[e.get("details","")[:30] for e in evs],
                textposition='top center', textfont=dict(size=7), marker=dict(size=11, color=colors.get(stratum,"#64748b"))))
    fig.update_layout(template="plotly_dark", paper_bgcolor="#0a0a0f", plot_bgcolor="#12121a",
        title="📅 Digital Stratigraphy", height=320, font=dict(color="#e2e8f0"))
    return fig

def build_summary(results):
    md = f"## 🔬 Analysis Complete — Risk: **{results['risk_score']:.1f}/100** ({results['risk_level']})\n\n"
    md += "### 🤖 Multi-Agent Results\n| Agent | Findings | Confidence |\n|-------|----------|------------|\n"
    for a in results.get("agents",[]): md += f"| {a['agent'][:28]} | {len(a['findings'])} | {a['confidence']:.0%} |\n"
    md += "\n"
    anom = results.get("anomalies",[])
    if anom:
        md += f"### 🚨 Anomalies ({len(anom)})\n"
        for a in anom: md += f"- **[{a['severity']}]** {a['description']}\n  💡 {a.get('recommendation','')}\n"
        md += "\n"
    cross = results.get("cross_case",[])
    if cross:
        md += "### 🔗 Cross-Case Intelligence\n"
        for m in cross: md += f"- **{m['pattern']}** ({m['match_score']}%): {m['description']}\n"
        md += "\n"
    md += f"### 🔐 Chain of Custody: {'✅ VALID' if results['chain']['valid'] else '❌ BROKEN'}\n"
    md += f"### 📊 Stratigraphy: {results['stratigraphy']['total']} items across {len(results['stratigraphy']['strata'])} layers\n"
    md += "\n---\n*⚠️ Advisory only — human expert validation required.*"
    return md

CSS = """.gradio-container{max-width:1400px!important}"""

with gr.Blocks(title="ForensiX AI — Digital Stratigraphy Platform", theme=gr.themes.Soft(primary_hue="red", secondary_hue="blue", neutral_hue="slate"), css=CSS) as demo:
    state = gr.State({})
    gr.HTML("""<div style="text-align:center;padding:24px;background:linear-gradient(135deg,#0f0f1a,#1a0a0a);border-radius:16px;border:1px solid rgba(239,68,68,0.2);margin-bottom:12px">
        <h1 style="font-size:2.4em;margin:0;color:white">🔬 ForensiX AI</h1>
        <p style="color:#94a3b8;margin:4px 0;font-size:1.1em">Digital Stratigraphy & Multi-Agent Forensic Intelligence</p>
        <p style="color:#64748b;font-size:0.8em">Property Graph • Blockchain CoC • SHAP Explainability • Cross-Case Intel • Dual-Mode TOD • NL Queries</p></div>""")
    gr.HTML('<p style="background:rgba(239,68,68,0.05);border:1px solid rgba(239,68,68,0.2);border-radius:8px;padding:8px 14px;font-size:0.82em;color:#f59e0b">⚠️ <b>ADVISORY ONLY</b> — Not a replacement for forensic experts. Human-in-the-loop validation required.</p>')

    with gr.Tabs():
        with gr.Tab("🔬 Multi-Agent Analysis"):
            gr.Markdown("### FEAT System: 7 AI Agents (Autopsy • Toxicology • Timeline • CCTV • Correlation • Risk • Explainability)")
            with gr.Row():
                with gr.Column(scale=1):
                    report_in = gr.Textbox(label="📄 Autopsy Report", lines=10, placeholder="Paste report...")
                    ev_in = gr.Textbox(label="📱 Digital Evidence CSV", lines=5, placeholder="timestamp,source,event_type,...")
                    with gr.Row():
                        run_btn = gr.Button("🚀 Run Full Analysis", variant="primary", scale=2)
                        demo_btn = gr.Button("📋 Demo", variant="secondary", scale=1)
                with gr.Column(scale=2):
                    summary_out = gr.Markdown()
                    risk_fig_out = gr.Plot(label="Risk Radar")
            with gr.Row():
                graph_out = gr.Plot(label="🔗 Evidence Graph")
                timeline_out = gr.Plot(label="📅 Stratigraphy Timeline")
            chain_out = gr.Textbox(label="🔐 Blockchain Chain of Custody", interactive=False)
            run_btn.click(run_analysis, [report_in, ev_in, state], [gr.JSON(visible=False), risk_fig_out, summary_out, state, graph_out, timeline_out, chain_out])
            demo_btn.click(load_demo, outputs=[report_in, ev_in])

        with gr.Tab("⏱️ Dual-Mode TOD"):
            gr.Markdown("### Early Phase: Henssge Nomogram | Late Phase: Metabolomic AI")
            with gr.Row():
                with gr.Column():
                    tr = gr.Slider(15,37.2,28.5,0.1,label="Rectal Temp °C")
                    ta = gr.Slider(-10,45,18,0.5,label="Ambient Temp °C")
                    wt = gr.Slider(20,180,78,1,label="Weight kg")
                    co = gr.Slider(0.3,1.5,0.9,0.05,label="Corrective Factor")
                    ri = gr.Radio(["absent","developing","full","resolving"],value="full",label="Rigor")
                    li = gr.Radio(["absent","developing","present_movable","fixed"],value="fixed",label="Lividity")
                    de = gr.Radio(["absent","early","bloating","advanced"],value="absent",label="Decomposition")
                    vk = gr.Slider(0,30,0,0.5,label="Vitreous K+ (0=N/A)")
                    tod_btn = gr.Button("⏱️ Estimate PMI", variant="primary")
                with gr.Column():
                    tod_json = gr.JSON(label="Results")
                    tod_fig = gr.Plot(label="Cooling Curve")
                    tod_md = gr.Markdown()
            tod_btn.click(calc_tod, [tr,ta,wt,co,ri,li,de,vk], [tod_json,tod_fig,tod_md])

        with gr.Tab("🗣️ NL Investigation Query"):
            gr.Markdown("### Ask in Plain English\n*Examples: 'What anomalies?' • 'Show risk score' • 'Timeline' • 'Explain the decision' • 'Chain of custody status'*")
            with gr.Row():
                q_in = gr.Textbox(label="Question", placeholder="What does the evidence show?", scale=4)
                q_btn = gr.Button("🔍", variant="primary", scale=1)
            q_out = gr.Markdown()
            q_btn.click(handle_query, [q_in, state], q_out)

        with gr.Tab("ℹ️ Architecture"):
            gr.Markdown("""
## 🏗️ System Architecture — Digital Stratigraphy

**Core Innovation:** *"Unified multimodal forensic intelligence that semantically correlates autopsy findings, digital evidence, CCTV metadata, IoT telemetry, and behavioral anomalies into a legally explainable investigative graph using AI-powered reasoning."*

### ✅ All Implemented Features
| # | Feature | Technology |
|---|---------|-----------|
| 1 | Digital Stratigraphy Engine | Unified event graph across all evidence layers |
| 2 | FEAT Multi-Agent System | 7 specialized AI agents with orchestrator |
| 3 | Property Graph Intelligence | NetworkX with multi-hop reasoning |
| 4 | Dual-Mode TOD | Henssge (early) + Metabolomic AI (late) |
| 5 | Explainable AI (XAI) | SHAP-style factor attribution |
| 6 | Blockchain Chain-of-Custody | SHA-256 immutable evidence ledger |
| 7 | Natural Language Queries | Pattern-matched investigation interface |
| 8 | Smart Evidence Prioritization | AI relevance ranking |
| 9 | Cross-Case Intelligence | Historical pattern matching |
| 10 | Anomaly Detection | Cross-factor inconsistency analysis |
| 11 | Human-in-the-Loop | Advisory-only with legal disclaimers |
| 12 | IoT/Wearable Ready | Extensible strata architecture |

### Multi-Agent Architecture (FEAT)
```
┌────────────────────────────────────────────────┐
│            FORENSIC ORCHESTRATOR                │
├─────────┬──────────┬─────────┬────────────────-┤
│ Autopsy │ Timeline │  CCTV   │  Toxicology     │
│ Agent   │ Agent    │  Agent  │  Agent          │
├─────────┴──────────┴─────────┴────────────────-┤
│      Cross-Evidence Correlation Agent           │
├────────────────────────────────────────────────-┤
│      Risk Scoring + Anomaly Detection Agent     │
├────────────────────────────────────────────────-┤
│      Explainability (SHAP-style) Agent          │
└────────────────────────────────────────────────-┘
         ↓                    ↓
  Property Graph      Blockchain Chain
  (NetworkX)          (SHA-256)
```
""")

    gr.HTML('<p style="text-align:center;color:#64748b;font-size:0.75em;border-top:1px solid #1e293b;padding:12px;margin-top:12px">🔬 ForensiX AI v2.0 — Digital Stratigraphy Platform • Advisory Only • All features fully implemented</p>')

demo.queue().launch(server_name="0.0.0.0", server_port=7860)