Massive update

2026-04-11 13:44:08 -04:00
parent fd6c2fd8bf
commit 3b3a136d04
31 changed files with 1919 additions and 94 deletions
@@ -0,0 +1,106 @@
+"""
+Rules-based intelligence extraction from call transcripts.
+
+Scans a transcript for known incident keywords, categorises the call, and
+extracts rough location hints (street/intersection mentions).
+
+No external ML dependencies — fast and always available even when STT is
+disabled. Designed to run as part of the post-upload background pipeline.
+"""
+import re
+from typing import Optional
+from app.internal.logger import logger
+from app.internal import firestore as fstore
+
+
+# ---------------------------------------------------------------------------
+# Keyword taxonomy
+# ---------------------------------------------------------------------------
+
+INCIDENT_KEYWORDS: dict[str, list[str]] = {
+    "fire": [
+        "fire", "smoke", "flames", "burning", "structure fire", "brush fire",
+        "wildfire", "arson", "working fire", "fully involved",
+    ],
+    "ems": [
+        "cardiac", "unconscious", "breathing", "overdose", "trauma",
+        "injury", "ambulance", "ems", "medic", "chest pain", "stroke",
+        "unresponsive", "fall", "laceration",
+    ],
+    "police": [
+        "pursuit", "chase", "shots fired", "weapon", "suspect", "robbery",
+        "assault", "burglary", "stolen", "fleeing", "armed", "shooting",
+        "stabbing", "domestic",
+    ],
+    "accident": [
+        "accident", "collision", "crash", "mvr", "vehicle", "rollover",
+        "hit and run", "ped", "pedestrian", "pi", "property damage",
+    ],
+}
+
+# Street suffix patterns for location extraction
+_STREET_RE = re.compile(
+    r'\b(?:\d+\s+)?[A-Z][a-zA-Z]+(?: [A-Z][a-zA-Z]+)*'
+    r'\s+(?:Street|St|Avenue|Ave|Boulevard|Blvd|Drive|Dr|Road|Rd|Lane|Ln'
+    r'|Court|Ct|Place|Pl|Way|Circle|Cir|Highway|Hwy|Route|Rt)\b',
+    re.IGNORECASE,
+)
+
+
+# ---------------------------------------------------------------------------
+# Public API
+# ---------------------------------------------------------------------------
+
+async def extract_tags(
+    call_id: str,
+    transcript: str,
+) -> tuple[list[str], Optional[str]]:
+    """
+    Extract incident tags from a transcript.
+
+    Returns:
+        (tags, primary_type) — e.g. (["fire", "structure fire"], "fire")
+        primary_type is the category with the most keyword hits, or None.
+
+    Side-effect: updates calls/{call_id}.tags in Firestore.
+    """
+    lower = transcript.lower()
+    matched: dict[str, list[str]] = {}
+
+    for category, keywords in INCIDENT_KEYWORDS.items():
+        hits = [kw for kw in keywords if kw in lower]
+        if hits:
+            matched[category] = hits
+
+    tags: list[str] = []
+    for category, hits in matched.items():
+        tags.append(category)
+        tags.extend(h for h in hits if h != category)
+
+    # Deduplicate while preserving order
+    seen: set[str] = set()
+    unique_tags: list[str] = []
+    for t in tags:
+        if t not in seen:
+            seen.add(t)
+            unique_tags.append(t)
+
+    # Primary type = category with most keyword hits
+    primary_type: Optional[str] = None
+    if matched:
+        primary_type = max(matched, key=lambda c: len(matched[c]))
+
+    if unique_tags:
+        try:
+            await fstore.doc_update("calls", call_id, {"tags": unique_tags})
+        except Exception as e:
+            logger.warning(f"Could not save tags for call {call_id}: {e}")
+
+    logger.info(f"Intelligence: call {call_id} → tags={unique_tags}, type={primary_type}")
+    return unique_tags, primary_type
+
+
+def extract_location_hint(transcript: str) -> Optional[str]:
+    """Return the first street-level location mention found in the transcript, or None."""
+    match = _STREET_RE.search(transcript)
+    return match.group(0) if match else None