Upload 6 files

.gitattributes

@@ -1,36 +1,4 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
 *.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
 *.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
-Data/Domain-A_Dataset_Clean.csv filter=lfs diff=lfs merge=lfs -text
+*.csv filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,7 @@
+__pycache__/
+*.py[cod]
+.DS_Store
+Thumbs.db
+.gradio/
+tmp/
+temp/

__init__.py

@@ -0,0 +1 @@
+"""Shared deployment utilities for the intelligent ticket auto-routing system."""

app.py

@@ -0,0 +1,853 @@
+"""
+Intelligent Ticket Auto-Routing System - Hugging Face Spaces App
+================================================================
+Converts support tickets into structured routing decisions:
+  - Multi-label tag classification
+  - Department routing (hybrid: tag-voting + semantic similarity)
+  - Priority prediction
+  - Duplicate detection via FAISS
+"""
+
+from __future__ import annotations
+
+import csv
+import os
+import tempfile
+import time
+import uuid
+from datetime import datetime
+from pathlib import Path
+
+import gradio as gr
+import joblib
+import numpy as np
+from sentence_transformers import SentenceTransformer
+
+try:
+    from .calibration_utils import (
+        calibrate_probabilities,
+        load_temperature_scaler,
+    )
+    from .duplicate_detection_utils import CachedDuplicateDetectionEngine
+    from .hybrid_routing_utils import (
+        DEFAULT_TAG_TO_DEPARTMENT,
+        assert_valid_routing_label_policy,
+        compute_department_hybrid_scores,
+        load_routing_label_policy,
+    )
+    from .review_policy_utils import (
+        apply_controlled_review,
+        load_review_policy,
+    )
+    from .runtime_utils import (
+        load_model_config,
+        load_routing_config,
+        resolve_dataset_file,
+        resolve_model_dir,
+        resolve_model_reference,
+    )
+except ImportError:  # pragma: no cover
+    from calibration_utils import (
+        calibrate_probabilities,
+        load_temperature_scaler,
+    )
+    from duplicate_detection_utils import CachedDuplicateDetectionEngine
+    from hybrid_routing_utils import (
+        DEFAULT_TAG_TO_DEPARTMENT,
+        assert_valid_routing_label_policy,
+        compute_department_hybrid_scores,
+        load_routing_label_policy,
+    )
+    from review_policy_utils import (
+        apply_controlled_review,
+        load_review_policy,
+    )
+    from runtime_utils import (
+        load_model_config,
+        load_routing_config,
+        resolve_dataset_file,
+        resolve_model_dir,
+        resolve_model_reference,
+    )
+
+
+APP_DIR = Path(__file__).resolve().parent
+MODEL_DIR = resolve_model_dir(APP_DIR)
+
+ROUTING_CONFIG, ROUTING_CONFIG_PATH = load_routing_config(APP_DIR)
+DEFAULT_DEPARTMENT = str(
+    ROUTING_CONFIG.get("default_department", "Human_Review")
+)
+PRIORITY_ESCALATION = {
+    str(priority).lower(): department
+    for priority, department in (ROUTING_CONFIG.get("priority_escalation") or {}).items()
+}
+LOG_PATH = os.path.join(tempfile.gettempdir(), "routing_evaluation_log.csv")
+
+
+print("Loading SBERT model...")
+model_config = load_model_config(APP_DIR)
+routing_sbert_model_name = resolve_model_reference(
+    model_config.get("sbert_model", "Eklavya73/sbert_finetuned"),
+    base_dir=APP_DIR,
+    model_dir=MODEL_DIR,
+)
+duplicate_sbert_model_name = resolve_model_reference(
+    model_config.get("duplicate_sbert_model", "Eklavya73/duplicate_sbert"),
+    base_dir=APP_DIR,
+    model_dir=MODEL_DIR,
+    default="all-mpnet-base-v2",
+)
+routing_sbert = SentenceTransformer(routing_sbert_model_name)
+duplicate_sbert = (
+    routing_sbert
+    if duplicate_sbert_model_name == routing_sbert_model_name
+    else SentenceTransformer(duplicate_sbert_model_name)
+)
+
+print("Loading classifiers...")
+tag_model = joblib.load(MODEL_DIR / "sbert_classifier.pkl")
+tag_calibrators = joblib.load(MODEL_DIR / "tag_calibrators.pkl")
+temperature_scaler = load_temperature_scaler(MODEL_DIR / "tag_temperature_scaler.pkl")
+review_policy = load_review_policy(MODEL_DIR / "routing_review_policy.pkl")
+
+priority_bundle = joblib.load(MODEL_DIR / "tuned_priority_model.pkl")
+priority_model = (
+    priority_bundle["model"]
+    if isinstance(priority_bundle, dict) and "model" in priority_bundle
+    else priority_bundle
+)
+priority_encoder = joblib.load(MODEL_DIR / "priority_encoder.pkl")
+hf_scaler = joblib.load(MODEL_DIR / "hf_scaler.pkl")
+
+tag_binarizer = joblib.load(MODEL_DIR / "mlb_tag_binarizer.pkl")
+tag_list = list(tag_binarizer.classes_)
+
+dept_prototypes = joblib.load(MODEL_DIR / "department_prototypes.pkl")
+routing_label_policy = load_routing_label_policy(
+    MODEL_DIR / "routing_label_policy.pkl",
+    fallback_tag_to_department=ROUTING_CONFIG.get(
+        "departments",
+        DEFAULT_TAG_TO_DEPARTMENT,
+    ),
+    valid_tags=tag_list,
+    valid_departments=dept_prototypes.keys(),
+    default_department=DEFAULT_DEPARTMENT,
+)
+tag_to_department = routing_label_policy["tag_to_department"]
+assert_valid_routing_label_policy(
+    routing_label_policy,
+    valid_tags=tag_list,
+    valid_departments=dept_prototypes.keys(),
+)
+
+print("Loading duplicate detection index...")
+duplicate_engine = CachedDuplicateDetectionEngine(APP_DIR)
+
+print(f"[OK] Tags: {len(tag_list)}, Departments: {len(dept_prototypes)}")
+print(f"[OK] Routing label policy: {len(tag_to_department)} active mappings")
+print(
+    "[OK] Routing config: "
+    f"{ROUTING_CONFIG_PATH if ROUTING_CONFIG_PATH is not None else 'defaults'}"
+)
+print(f"[OK] Default human-review department: {DEFAULT_DEPARTMENT}")
+print(f"[OK] Routing SBERT model: {routing_sbert_model_name}")
+print(f"[OK] Duplicate SBERT model: {duplicate_sbert_model_name}")
+print(f"[OK] Duplicate index: {duplicate_engine.index_size} vectors")
+print(f"[OK] Duplicate threshold: {duplicate_engine.duplicate_threshold:.4f}")
+print(f"[OK] Temperature scaler: T={temperature_scaler.get('temperature', 1.0):.3f}")
+print(
+    "[OK] Review policy: "
+    f"target={review_policy.get('target_review_fraction', 0.15):.0%}, "
+    f"percentile_threshold={review_policy.get('percentile_threshold', 0.55):.3f}, "
+    f"fallback_threshold={review_policy.get('fallback_threshold', 0.55):.3f}"
+)
+
+
+def encode_ticket_embedding(text, encoder):
+    emb = np.asarray(encoder.encode(text), dtype=float).reshape(-1)
+    emb_norm = np.linalg.norm(emb)
+    if emb_norm == 0.0:
+        return emb
+    return emb / emb_norm
+
+
+def predict_tags(text, emb):
+    raw_probs = np.asarray(tag_model.predict_proba([emb])[0], dtype=float)
+    calibrated = calibrate_probabilities(
+        raw_probs,
+        tag_calibrators=tag_calibrators,
+        temperature_scaler=temperature_scaler,
+    )
+    top_idx = calibrated.argsort()[-5:][::-1]
+    return top_idx, calibrated[top_idx], calibrated, raw_probs
+
+
+def extract_features(text):
+    words = text.split()
+    return [
+        len(text),
+        len(words),
+        len(set(words)) / (len(words) + 1),
+        np.mean([len(word) for word in words]) if words else 0,
+        sum(word in text.lower() for word in ["urgent", "critical", "down"]),
+        sum(word in text.lower() for word in ["not", "cannot", "no"]),
+    ]
+
+
+def predict_priority(text, emb, return_confidence=False):
+    features = extract_features(text)
+    features_scaled = hf_scaler.transform([features])
+    x = np.hstack([emb.reshape(1, -1), features_scaled])
+    pred_idx = int(priority_model.predict(x)[0])
+    priority_label = str(priority_encoder.classes_[pred_idx])
+    priority_confidence = float("nan")
+
+    if hasattr(priority_model, "predict_proba"):
+        try:
+            probs = np.asarray(
+                priority_model.predict_proba(x)[0],
+                dtype=float,
+            ).reshape(-1)
+            if probs.size:
+                priority_confidence = float(probs[pred_idx])
+        except Exception:
+            priority_confidence = float("nan")
+
+    if return_confidence:
+        return priority_label, priority_confidence
+    return priority_label
+
+
+HYBRID_CLASSIFIER_WEIGHT = 0.7
+HYBRID_SIMILARITY_WEIGHT = 0.3
+HYBRID_FLOOR = 0.45
+FLAGGED_HYBRID_FLOOR = 0.30
+MARGIN_THRESHOLD = 0.15
+ENTROPY_THRESHOLD = 1.8
+
+
+def compute_confidence_metrics(calibrated_probs):
+    probs = np.asarray(calibrated_probs, dtype=float).reshape(-1)
+    if probs.size == 0:
+        return 0.0, float("inf")
+
+    sorted_probs = np.sort(probs)[::-1]
+    top1 = float(sorted_probs[0])
+    top2 = float(sorted_probs[1]) if len(sorted_probs) > 1 else 0.0
+    margin = top1 - top2
+
+    p = np.clip(probs, 1e-12, None)
+    total = float(p.sum())
+    if total == 0.0:
+        p = np.full_like(p, 1.0 / len(p))
+    else:
+        p = p / total
+    entropy = float(-np.sum(p * np.log(p)))
+    return margin, entropy
+
+
+def decide_routing_mode(hybrid_confidence, calibrated_probs):
+    margin, entropy = compute_confidence_metrics(calibrated_probs)
+
+    if hybrid_confidence < HYBRID_FLOOR:
+        return "HUMAN_REVIEW", True, margin, entropy
+
+    if (margin >= MARGIN_THRESHOLD) or (entropy <= ENTROPY_THRESHOLD):
+        return "AUTO_ROUTE", False, margin, entropy
+
+    if hybrid_confidence >= FLAGGED_HYBRID_FLOOR:
+        return "AUTO_ROUTE_FLAGGED", True, margin, entropy
+
+    return "HUMAN_REVIEW", True, margin, entropy
+
+
+def route_ticket(emb, text):
+    _, _, calibrated_probs, _ = predict_tags(text, emb)
+    best_dept, hybrid_confidence, department_details, top_tag_votes = (
+        compute_department_hybrid_scores(
+            calibrated_probs,
+            emb,
+            dept_prototypes,
+            tag_to_department=tag_to_department,
+            tag_names=tag_list,
+            classifier_weight=HYBRID_CLASSIFIER_WEIGHT,
+            similarity_weight=HYBRID_SIMILARITY_WEIGHT,
+            top_k=5,
+        )
+    )
+    priority, priority_confidence = predict_priority(
+        text,
+        emb,
+        return_confidence=True,
+    )
+    base_mode, _, margin, entropy = decide_routing_mode(
+        hybrid_confidence,
+        calibrated_probs,
+    )
+
+    recommended_department = best_dept
+    routed_department = recommended_department
+    escalation_note = ""
+
+    if not top_tag_votes or best_dept is None:
+        review_decision = {
+            "base_mode": "HUMAN_REVIEW",
+            "final_mode": "HUMAN_REVIEW",
+            "forced_human_review": False,
+            "percentile_threshold": float(
+                review_policy.get("percentile_threshold", 0.55)
+            ),
+            "fallback_threshold": float(
+                review_policy.get("fallback_threshold", 0.55)
+            ),
+            "reason": "No valid tag votes or department resolved. Requires human review.",
+        }
+        return {
+            "mode": "HUMAN_REVIEW",
+            "department": DEFAULT_DEPARTMENT,
+            "recommended_department": None,
+            "priority": priority,
+            "priority_confidence": priority_confidence,
+            "hybrid_confidence": hybrid_confidence,
+            "review": True,
+            "margin": margin,
+            "entropy": entropy,
+            "best_details": {},
+            "top_tag_votes": [],
+            "review_decision": review_decision,
+            "note": review_decision["reason"],
+        }
+
+    escalation_department = PRIORITY_ESCALATION.get(str(priority).lower())
+    if base_mode != "HUMAN_REVIEW" and escalation_department:
+        routed_department = str(escalation_department)
+        escalation_note = (
+            f" Priority escalation override applied after gate: "
+            f"{priority} -> {routed_department}."
+        )
+
+    mode, review, review_decision = apply_controlled_review(
+        base_mode,
+        hybrid_confidence,
+        review_policy=review_policy,
+    )
+
+    if review_decision.get("forced_human_review", False):
+        final_department = DEFAULT_DEPARTMENT
+        note = (
+            f"{review_decision.get('reason', '')} "
+            f"Recommended department before override: {routed_department}."
+            f"{escalation_note}"
+        ).strip()
+    elif mode == "AUTO_ROUTE":
+        final_department = routed_department
+        note = (
+            f"Stage 2 pass: hybrid_confidence={hybrid_confidence:.4f}, "
+            f"margin={margin:.4f}, entropy={entropy:.4f}."
+            f"{escalation_note}"
+        )
+    elif mode == "AUTO_ROUTE_FLAGGED":
+        final_department = routed_department
+        note = (
+            f"Stage 2 flagged: hybrid_confidence={hybrid_confidence:.4f}, "
+            f"margin={margin:.4f}, entropy={entropy:.4f}."
+            f"{escalation_note}"
+        )
+    elif hybrid_confidence < HYBRID_FLOOR:
+        final_department = DEFAULT_DEPARTMENT
+        note = (
+            f"Stage 1 reject: hybrid_confidence {hybrid_confidence:.4f} "
+            f"< HYBRID_FLOOR {HYBRID_FLOOR}."
+        )
+    else:
+        final_department = DEFAULT_DEPARTMENT
+        note = (
+            f"Stage 2 reject: hybrid_confidence={hybrid_confidence:.4f}, "
+            f"margin={margin:.4f}, entropy={entropy:.4f}."
+        )
+
+    best_details = department_details.get(recommended_department, {})
+    return {
+        "mode": mode,
+        "department": final_department,
+        "recommended_department": recommended_department,
+        "priority": priority,
+        "priority_confidence": priority_confidence,
+        "hybrid_confidence": hybrid_confidence,
+        "review": review,
+        "margin": margin,
+        "entropy": entropy,
+        "best_details": best_details,
+        "top_tag_votes": top_tag_votes,
+        "review_decision": review_decision,
+        "note": note.strip(),
+    }
+
+
+LOG_COLUMNS = [
+    "ticket_id",
+    "timestamp",
+    "ticket_text",
+    "duplicate_flag",
+    "duplicate_score",
+    "routing_mode",
+    "department",
+    "base_routing_mode",
+    "requires_review",
+    "controlled_review_applied",
+    "department_confidence",
+    "classifier_confidence",
+    "semantic_similarity",
+    "raw_semantic_similarity",
+    "priority",
+    "priority_confidence",
+    "selected_tags",
+    "routing_score",
+    "hybrid_confidence",
+    "margin",
+    "entropy",
+    "review_percentile_threshold",
+    "review_fallback_threshold",
+    "prediction_latency_ms",
+    "explanation",
+]
+
+
+def _ensure_log_header():
+    if not os.path.exists(LOG_PATH):
+        with open(LOG_PATH, "w", newline="", encoding="utf-8") as handle:
+            csv.writer(handle).writerow(LOG_COLUMNS)
+        return
+
+    with open(LOG_PATH, "r", newline="", encoding="utf-8") as handle:
+        existing_header = next(csv.reader(handle), [])
+
+    if existing_header != LOG_COLUMNS:
+        with open(LOG_PATH, "w", newline="", encoding="utf-8") as handle:
+            csv.writer(handle).writerow(LOG_COLUMNS)
+
+
+def _append_log(row_dict):
+    _ensure_log_header()
+    with open(LOG_PATH, "a", newline="", encoding="utf-8") as handle:
+        csv.writer(handle).writerow([row_dict.get(column, "") for column in LOG_COLUMNS])
+
+
+def process_ticket(text):
+    t0 = time.time()
+    ticket_id = str(uuid.uuid4())[:8]
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+
+    routing_emb = encode_ticket_embedding(text, routing_sbert)
+    duplicate_emb = encode_ticket_embedding(text, duplicate_sbert)
+
+    best_match = duplicate_engine.find_best_match(duplicate_emb, k=20)
+    dup_score = (
+        float(best_match["similarity"])
+        if best_match is not None
+        else 0.0
+    )
+    dup_text = best_match.get("matched_text") if best_match is not None else None
+    is_dup = bool(
+        best_match is not None
+        and dup_score >= float(duplicate_engine.duplicate_threshold)
+    )
+
+    routing = route_ticket(routing_emb, text)
+    latency_ms = round((time.time() - t0) * 1000, 2)
+
+    mode = routing["mode"]
+    dept = routing["department"]
+    priority = routing["priority"]
+    priority_confidence = routing["priority_confidence"]
+    hybrid_confidence = routing["hybrid_confidence"]
+    review = routing["review"]
+    margin = routing["margin"]
+    entropy = routing["entropy"]
+    best_details = routing["best_details"]
+    top_tag_votes = routing["top_tag_votes"]
+    review_decision = routing["review_decision"]
+    note = routing["note"]
+
+    classifier_confidence = float(best_details.get("classifier_confidence", 0.0))
+    semantic_similarity = float(best_details.get("semantic_similarity", 0.0))
+    raw_semantic_similarity = float(best_details.get("raw_semantic_similarity", 0.0))
+    base_mode = str(review_decision.get("base_mode", mode))
+    review_reason = str(review_decision.get("reason", note))
+    percentile_threshold = float(
+        review_decision.get(
+            "percentile_threshold",
+            review_policy.get("percentile_threshold", 0.55),
+        )
+    )
+    fallback_threshold = float(
+        review_decision.get(
+            "fallback_threshold",
+            review_policy.get("fallback_threshold", 0.55),
+        )
+    )
+    controlled_review_applied = bool(
+        review_decision.get("forced_human_review", False)
+    )
+    recommended_department = routing.get("recommended_department")
+    tag_summary = ", ".join(
+        f"{vote['tag']} ({vote['score']:.2f})"
+        for vote in top_tag_votes[:3]
+    )
+
+    recommended_text = (
+        f" Recommended department before final policy: {recommended_department}."
+        if recommended_department and recommended_department != dept
+        else ""
+    )
+
+    if is_dup:
+        explanation = (
+            f"Duplicate detected (score={dup_score:.4f}). "
+            f"Original: {str(dup_text)[:100]}. "
+            f"Routing mode: {mode} (base_mode={base_mode}), "
+            f"final_department={dept}, hybrid_confidence={hybrid_confidence:.3f}, "
+            f"classifier_confidence={classifier_confidence:.3f}, "
+            f"semantic_similarity={semantic_similarity:.3f} "
+            f"(raw={raw_semantic_similarity:.3f}), margin={margin:.3f}, "
+            f"entropy={entropy:.3f}, controlled_review_applied={controlled_review_applied}, "
+            f"review_thresholds=(percentile={percentile_threshold:.3f}, "
+            f"fallback={fallback_threshold:.3f}).{recommended_text} {note}"
+        )
+        result = {
+            "ticket_id": ticket_id,
+            "status": "DUPLICATE",
+            "route": mode,
+            "department": dept,
+            "priority": priority,
+            "confidence": round(float(hybrid_confidence), 3),
+            "review": review,
+            "tags": tag_summary,
+            "message": (
+                f"Duplicate of: {str(dup_text)[:200]} (similarity={dup_score:.3f}). "
+                f"{note}"
+            ).strip(),
+            "latency": latency_ms,
+        }
+    else:
+        explanation = (
+            f"Ticket processed with final department {dept}. "
+            f"Predicted tags [{tag_summary}] produced routing mode {mode} "
+            f"(base_mode={base_mode}), hybrid_confidence={hybrid_confidence:.3f}, "
+            f"classifier_confidence={classifier_confidence:.3f}, "
+            f"semantic_similarity={semantic_similarity:.3f} "
+            f"(raw={raw_semantic_similarity:.3f}), margin={margin:.3f}, "
+            f"entropy={entropy:.3f}, controlled_review_applied={controlled_review_applied}, "
+            f"review_thresholds=(percentile={percentile_threshold:.3f}, "
+            f"fallback={fallback_threshold:.3f}).{recommended_text} {review_reason}"
+        )
+        result = {
+            "ticket_id": ticket_id,
+            "status": "NOT DUPLICATE",
+            "route": mode,
+            "department": dept,
+            "priority": priority,
+            "confidence": round(float(hybrid_confidence), 3),
+            "review": review,
+            "tags": tag_summary,
+            "message": note if note else "Ticket processed successfully",
+            "latency": latency_ms,
+        }
+
+    duplicate_engine.add_ticket(ticket_id, text, embedding=duplicate_emb)
+    _append_log(
+        {
+            "ticket_id": ticket_id,
+            "timestamp": timestamp,
+            "ticket_text": text,
+            "duplicate_flag": is_dup,
+            "duplicate_score": round(float(dup_score), 4),
+            "routing_mode": mode,
+            "department": dept,
+            "department_confidence": round(float(hybrid_confidence), 4),
+            "base_routing_mode": base_mode,
+            "requires_review": bool(review),
+            "controlled_review_applied": controlled_review_applied,
+            "classifier_confidence": round(float(classifier_confidence), 4),
+            "semantic_similarity": round(float(semantic_similarity), 4),
+            "raw_semantic_similarity": round(float(raw_semantic_similarity), 4),
+            "priority": priority,
+            "priority_confidence": (
+                round(float(priority_confidence), 4)
+                if np.isfinite(priority_confidence)
+                else ""
+            ),
+            "selected_tags": tag_summary,
+            "routing_score": round(float(hybrid_confidence), 4),
+            "hybrid_confidence": round(float(hybrid_confidence), 4),
+            "margin": round(float(margin), 4),
+            "entropy": round(float(entropy), 4),
+            "review_percentile_threshold": round(float(percentile_threshold), 4),
+            "review_fallback_threshold": round(float(fallback_threshold), 4),
+            "prediction_latency_ms": latency_ms,
+            "explanation": explanation,
+        }
+    )
+
+    return result
+
+
+def ui_process(text):
+    if not text or not text.strip():
+        return ("Please enter ticket text", "", "", "", "", "", "", "", "")
+
+    result = process_ticket(text.strip())
+    conf_pct = int(result["confidence"] * 100)
+
+    if result["route"] == "HUMAN_REVIEW":
+        review_badge = "Human review required"
+    elif result["route"] == "AUTO_ROUTE_FLAGGED":
+        review_badge = "QA review required"
+    else:
+        review_badge = "No"
+
+    priority_map = {
+        "critical": "Critical",
+        "high": "High",
+        "medium": "Medium",
+        "low": "Low",
+    }
+    priority_display = priority_map.get(
+        result["priority"].lower(),
+        result["priority"],
+    )
+
+    route_map = {
+        "AUTO_ROUTE": "Auto-Routed",
+        "AUTO_ROUTE_FLAGGED": "Auto-Routed + Flagged",
+        "HUMAN_REVIEW": "Human Review Required",
+    }
+    route_display = route_map.get(result["route"], result["route"])
+    dept_display = result["department"].replace("_", " ")
+
+    return (
+        result["status"],
+        result["ticket_id"],
+        route_display,
+        dept_display,
+        priority_display,
+        f"{conf_pct}%",
+        result["tags"],
+        review_badge,
+        result["message"],
+    )
+
+
+CSS = """
+@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap');
+
+* { font-family: 'Inter', sans-serif !important; }
+
+.gradio-container {
+    max-width: 960px !important;
+    margin: 0 auto !important;
+}
+
+.app-header {
+    text-align: center;
+    padding: 1.5rem 1rem;
+    background: linear-gradient(135deg, #4f46e5 0%, #7c3aed 50%, #a855f7 100%);
+    border-radius: 16px;
+    margin-bottom: 1.5rem;
+    box-shadow: 0 8px 32px rgba(79, 70, 229, 0.3);
+}
+.app-header h1 {
+    color: white !important;
+    font-size: 1.75rem !important;
+    font-weight: 700 !important;
+    margin: 0 !important;
+    letter-spacing: -0.02em;
+}
+.app-header p {
+    color: rgba(255,255,255,0.85) !important;
+    font-size: 0.95rem !important;
+    margin: 0.4rem 0 0 0 !important;
+}
+
+.result-card {
+    background: linear-gradient(145deg, rgba(255,255,255,0.05), rgba(255,255,255,0.02));
+    border: 1px solid rgba(255,255,255,0.1);
+    border-radius: 12px;
+    padding: 0.25rem;
+}
+
+.status-box textarea, .status-box input {
+    font-weight: 600 !important;
+    font-size: 1rem !important;
+}
+
+.submit-btn {
+    background: linear-gradient(135deg, #4f46e5, #7c3aed) !important;
+    border: none !important;
+    color: white !important;
+    font-weight: 600 !important;
+    font-size: 1rem !important;
+    padding: 0.75rem 2rem !important;
+    border-radius: 10px !important;
+    box-shadow: 0 4px 16px rgba(79, 70, 229, 0.4) !important;
+    transition: all 0.3s ease !important;
+}
+.submit-btn:hover {
+    transform: translateY(-2px) !important;
+    box-shadow: 0 6px 24px rgba(79, 70, 229, 0.5) !important;
+}
+
+.clear-btn {
+    border: 1px solid rgba(255,255,255,0.2) !important;
+    border-radius: 10px !important;
+    font-weight: 500 !important;
+}
+
+.stats-row {
+    text-align: center;
+    padding: 0.75rem;
+    background: rgba(79, 70, 229, 0.08);
+    border-radius: 10px;
+    margin-top: 0.5rem;
+    font-size: 0.85rem;
+    color: #a5b4fc;
+}
+
+footer { display: none !important; }
+"""
+
+
+EXAMPLES = [
+    [
+        "My laptop screen is flickering and sometimes goes completely black. "
+        "I've tried restarting but the issue persists after login."
+    ],
+    [
+        "I cannot access the company VPN from my home network. It keeps showing "
+        "authentication failed error even though my password is correct."
+    ],
+    [
+        "We need to upgrade our database server as the current one is running out "
+        "of storage space and response times have increased significantly."
+    ],
+    [
+        "I was charged twice for my last month's subscription. Please process a "
+        "refund for the duplicate charge."
+    ],
+    [
+        "The email server has been down since this morning. No one in the office "
+        "can send or receive emails. This is critical!"
+    ],
+    [
+        "Can you provide training materials for the new CRM software that was "
+        "deployed last week?"
+    ],
+]
+
+
+with gr.Blocks(
+    css=CSS,
+    theme=gr.themes.Soft(primary_hue="indigo", neutral_hue="slate"),
+    title="Ticket Auto-Routing System",
+) as app:
+    gr.HTML(
+        """
+        <div class="app-header">
+            <h1>Intelligent Ticket Auto-Routing System</h1>
+            <p>AI-powered ticket classification, routing, priority prediction and duplicate detection</p>
+        </div>
+        """
+    )
+
+    with gr.Row():
+        with gr.Column(scale=1):
+            ticket_input = gr.Textbox(
+                label="Ticket Description",
+                placeholder="Describe the support issue in detail...",
+                lines=6,
+                max_lines=12,
+            )
+            with gr.Row():
+                submit_btn = gr.Button(
+                    "Process Ticket",
+                    variant="primary",
+                    elem_classes=["submit-btn"],
+                )
+                clear_btn = gr.ClearButton(
+                    value="Clear",
+                    elem_classes=["clear-btn"],
+                )
+
+            gr.Examples(
+                examples=EXAMPLES,
+                inputs=ticket_input,
+                label="Try these examples",
+            )
+
+        with gr.Column(scale=1):
+            with gr.Group(elem_classes=["result-card"]):
+                dup_status = gr.Textbox(
+                    label="Duplicate Status",
+                    interactive=False,
+                    elem_classes=["status-box"],
+                )
+                ticket_id = gr.Textbox(label="Ticket ID", interactive=False)
+
+            with gr.Group(elem_classes=["result-card"]):
+                with gr.Row():
+                    route_mode = gr.Textbox(
+                        label="Routing Mode",
+                        interactive=False,
+                    )
+                    department = gr.Textbox(
+                        label="Department",
+                        interactive=False,
+                    )
+                with gr.Row():
+                    priority = gr.Textbox(label="Priority", interactive=False)
+                    confidence = gr.Textbox(
+                        label="Hybrid Confidence",
+                        interactive=False,
+                    )
+
+            with gr.Group(elem_classes=["result-card"]):
+                tags = gr.Textbox(label="Predicted Tags", interactive=False)
+                needs_review = gr.Textbox(label="Needs Review", interactive=False)
+                message = gr.Textbox(
+                    label="Details",
+                    interactive=False,
+                    lines=2,
+                )
+
+    gr.HTML(
+        f"""
+        <div class="stats-row">
+            Database: <strong>{duplicate_engine.index_size:,}</strong> tickets indexed
+            &nbsp;|&nbsp;
+            <strong>{len(tag_list)}</strong> tag categories
+            &nbsp;|&nbsp;
+            <strong>{len(dept_prototypes)}</strong> departments
+        </div>
+        """
+    )
+
+    outputs = [
+        dup_status,
+        ticket_id,
+        route_mode,
+        department,
+        priority,
+        confidence,
+        tags,
+        needs_review,
+        message,
+    ]
+
+    submit_btn.click(fn=ui_process, inputs=ticket_input, outputs=outputs)
+    ticket_input.submit(fn=ui_process, inputs=ticket_input, outputs=outputs)
+    clear_btn.add([ticket_input] + outputs)
+
+
+if __name__ == "__main__":
+    app.launch()

requirements.txt

@@ -0,0 +1,14 @@
+gradio==5.23.0
+sentence-transformers==5.2.3
+transformers==4.49.0
+torch==2.6.0
+faiss-cpu==1.13.2
+scikit-learn==1.5.1
+scipy==1.13.1
+numpy==1.26.4
+pandas==2.2.3
+joblib==1.4.2
+PyYAML==6.0.2
+xgboost==3.2.0
+lightgbm==4.6.0
+huggingface_hub==0.16.4

runtime.txt

@@ -0,0 +1 @@
+python-3.10.16