Add LLM-based patent classification tagging by technology domain

- Add classify_patent_tags() to LLMAnalyzer with canonical tag list
  (ai, semiconductors, materials, biotech, networking, other)
- Add patent_tags TEXT[] column to patents table with GIN index
- Run classification automatically in the analysis pipeline after
  patent processing; persist tags via update_patent_tags()
- Include tags in CompanyAnalysisResult and API response models
- Add ?tags= filter to GET /analyze/batch endpoint
- Add GET /analytics/tags endpoint for tag distribution data
- Add tag filter controls and distribution chart to Analytics page
- Add 12 unit tests covering classification, DB storage, and caching

Closes leeworks-agents/SPARC#1672

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

SPARC/analyzer.py

@@ -40,8 +40,9 @@ class CompanyAnalyzer:
         1. Retrieve patents from SERP API
         2. Download and parse each patent PDF
         3. Minimize patent content (remove bloat)
-        4. Analyze portfolio with LLM
-        5. Return performance estimation
+        4. Classify patent technology domain tags
+        5. Analyze portfolio with LLM
+        6. Return performance estimation
 
         Args:
           company_name: Name of the company to analyze
@@ -97,6 +98,17 @@ class CompanyAnalyzer:
         if not processed_patents:
             return f"Failed to process any patents for {company_name}"
 
+        # Classify each patent's technology domain tags
+        logger.info("Classifying patent technology domains...")
+        for patent_data in processed_patents:
+            if "tags" not in patent_data or not patent_data["tags"]:
+                tags = self.llm_analyzer.classify_patent_tags(
+                    patent_content=patent_data["content"], model=model
+                )
+                patent_data["tags"] = tags
+                # Persist tags to the database
+                self.db.update_patent_tags(patent_data["patent_id"], tags)
+
         logger.info("Analyzing portfolio with LLM...")
 
         # Analyze the full portfolio with LLM
@@ -181,7 +193,10 @@ class CompanyAnalyzer:
             if db:
                 cached = db.get_cached_patent(patent.patent_id)
                 if cached and cached.get("minimized_content"):
-                    return {"patent_id": patent.patent_id, "content": cached["minimized_content"]}
+                    result = {"patent_id": patent.patent_id, "content": cached["minimized_content"]}
+                    if cached.get("patent_tags"):
+                        result["tags"] = cached["patent_tags"]
+                    return result
 
             # Full processing: download, parse, minimize
             patent = SERP.save_patents(patent)
@@ -217,11 +232,19 @@ class CompanyAnalyzer:
             # Delegate to analyze_company which handles SERP/patent caching
             analysis = self.analyze_company(company_name, model=model)
 
-            # Determine patent count from cached SERP query
+            # Determine patent count and aggregate tags from cached SERP query
             query_hash = hashlib.sha256(company_name.lower().encode()).hexdigest()
             cached_ids = self.db.get_cached_serp_query(query_hash)
             patent_count = len(cached_ids) if cached_ids else 0
 
+            # Collect unique tags across all patents for this company
+            all_tags: set[str] = set()
+            if cached_ids:
+                for pid in cached_ids:
+                    cached_patent = self.db.get_cached_patent(pid)
+                    if cached_patent and cached_patent.get("patent_tags"):
+                        all_tags.update(cached_patent["patent_tags"])
+
             # Check if analysis indicates failure
             if analysis.startswith("No patents found") or analysis.startswith(
                 "Failed to process"
@@ -239,6 +262,7 @@ class CompanyAnalyzer:
                 analysis=analysis,
                 patent_count=patent_count,
                 success=True,
+                tags=sorted(all_tags),
             )
 
         except Exception as e:

SPARC/api.py

@@ -57,6 +57,7 @@ class CompanyAnalysisResponse(BaseModel):
     success: bool
     error: str | None = None
     model: str | None = None
+    tags: list[str] = []
     timestamp: datetime
 
 
@@ -188,6 +189,7 @@ def _convert_result(result: CompanyAnalysisResult) -> CompanyAnalysisResponse:
         success=result.success,
         error=result.error,
         model=result.model,
+        tags=result.tags,
         timestamp=result.timestamp,
     )
 
@@ -560,6 +562,38 @@ async def get_analytics(
     )
 
 
+@app.get("/analytics/tags", tags=["Analytics"])
+async def get_tag_distribution(
+    _: UserResponse = Depends(get_current_user),
+):
+    """Get the distribution of technology domain tags across all patents.
+
+    Returns:
+        List of tag counts and the canonical tag list
+    """
+    from SPARC.llm import LLMAnalyzer
+    db = get_db_client()
+
+    with db.get_conn() as conn:
+        with conn.cursor() as cur:
+            cur.execute(
+                """
+                SELECT tag, COUNT(*) as count
+                FROM patents, UNNEST(patent_tags) AS tag
+                WHERE patent_tags IS NOT NULL
+                GROUP BY tag
+                ORDER BY count DESC
+                """
+            )
+            rows = cur.fetchall()
+
+    by_tag = [{"tag": row[0], "count": row[1]} for row in rows]
+    return {
+        "by_tag": by_tag,
+        "canonical_tags": LLMAnalyzer.CANONICAL_TAGS,
+    }
+
+
 # ============== Model Selection Endpoints ==============
 
 # Supported models via OpenRouter
@@ -969,6 +1003,10 @@ async def list_analysis_results(
         str | None,
         Query(description="Filter results by company name"),
     ] = None,
+    tags: Annotated[
+        str | None,
+        Query(description="Comma-separated technology domain tags to filter by (e.g. 'ai,semiconductors')"),
+    ] = None,
     limit: Annotated[int, Query(ge=1, le=200)] = 50,
     cursor: Annotated[
         str | None,
@@ -986,14 +1024,27 @@ async def list_analysis_results(
 
     Args:
         company_name: Optional filter by company name
+        tags: Optional comma-separated tag filter (e.g. 'ai,semiconductors')
         limit: Maximum number of results to return (default 50, max 200)
         cursor: Opaque pagination cursor from a previous response
 
     Returns:
         Paginated list of analysis results
     """
+    # Parse and validate tags
+    tag_list = None
+    if tags:
+        from SPARC.llm import LLMAnalyzer
+        tag_list = [t.strip().lower() for t in tags.split(",") if t.strip()]
+        invalid = [t for t in tag_list if t not in LLMAnalyzer.CANONICAL_TAGS]
+        if invalid:
+            raise HTTPException(
+                status_code=400,
+                detail=f"Invalid tags: {', '.join(invalid)}. Valid tags: {', '.join(LLMAnalyzer.CANONICAL_TAGS)}",
+            )
+
     db = _get_job_db()
-    rows = db.list_analyses(company_name=company_name, limit=limit + 1, cursor=cursor)
+    rows = db.list_analyses(company_name=company_name, limit=limit + 1, cursor=cursor, tags=tag_list)
 
     has_next = len(rows) > limit
     if has_next:

SPARC/database.py

@@ -146,15 +146,35 @@ class DatabaseClient:
                     pdf_link TEXT,
                     raw_sections JSONB,
                     minimized_content TEXT,
+                    patent_tags TEXT[],
                     created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
                 )
             """)
 
+            # Add patent_tags column if it doesn't exist (for existing tables)
+            cursor.execute("""
+                DO $$
+                BEGIN
+                    IF NOT EXISTS (
+                        SELECT 1 FROM information_schema.columns
+                        WHERE table_name = 'patents' AND column_name = 'patent_tags'
+                    ) THEN
+                        ALTER TABLE patents ADD COLUMN patent_tags TEXT[];
+                    END IF;
+                END $$;
+            """)
+
             cursor.execute("""
                 CREATE INDEX IF NOT EXISTS idx_patents_company
                 ON patents(company_name)
             """)
 
+            # GIN index for efficient tag array queries
+            cursor.execute("""
+                CREATE INDEX IF NOT EXISTS idx_patents_tags
+                ON patents USING GIN(patent_tags)
+            """)
+
             # Create SERP query cache table
             cursor.execute("""
                 CREATE TABLE IF NOT EXISTS serp_queries (
@@ -376,6 +396,7 @@ class DatabaseClient:
         company_name: Optional[str] = None,
         limit: int = 50,
         cursor: Optional[str] = None,
+        tags: Optional[List[str]] = None,
     ) -> List[Dict]:
         """List analysis results with cursor-based pagination.
 
@@ -383,29 +404,40 @@ class DatabaseClient:
             company_name: Optional filter by company name.
             limit: Maximum number of records to return.
             cursor: Opaque cursor (``timestamp|id``) from a previous response.
+            tags: Optional list of technology domain tags to filter by.
 
         Returns:
             List of analysis dicts ordered by timestamp descending.
         """
-        conditions: list[str] = ["is_cached = FALSE"]
+        conditions: list[str] = ["m.is_cached = FALSE"]
         params: list = []
+        join_clause = ""
 
         if company_name:
-            conditions.append("LOWER(company_name) = LOWER(%s)")
+            conditions.append("LOWER(m.company_name) = LOWER(%s)")
             params.append(company_name)
 
+        if tags:
+            # Join with patents table to filter by tags
+            join_clause = (
+                " INNER JOIN patents p ON LOWER(p.company_name) = LOWER(m.company_name)"
+            )
+            conditions.append("p.patent_tags && %s")
+            params.append(tags)
+
         if cursor:
             try:
                 ts_str, cursor_id = cursor.rsplit("|", 1)
-                conditions.append("(timestamp, id) < (%s, %s)")
+                conditions.append("(m.timestamp, m.id) < (%s, %s)")
                 params.extend([ts_str, int(cursor_id)])
             except (ValueError, TypeError):
                 pass  # Ignore malformed cursors; return from start
 
-        query = "SELECT id, company_name, analysis_type, model, response, timestamp FROM llm_messages"
+        query = "SELECT DISTINCT m.id, m.company_name, m.analysis_type, m.model, m.response, m.timestamp FROM llm_messages m"
+        query += join_clause
         if conditions:
             query += " WHERE " + " AND ".join(conditions)
-        query += " ORDER BY timestamp DESC, id DESC LIMIT %s"
+        query += " ORDER BY m.timestamp DESC, m.id DESC LIMIT %s"
         params.append(limit)
 
         with self.get_conn() as conn:
@@ -493,22 +525,63 @@ class DatabaseClient:
         pdf_link: str,
         raw_sections: Dict,
         minimized_content: str,
+        patent_tags: Optional[List[str]] = None,
     ) -> None:
         """Store a processed patent in the cache."""
         with self.get_conn() as conn:
             with conn.cursor() as cursor:
                 cursor.execute(
                     """
-                    INSERT INTO patents (patent_id, company_name, pdf_link, raw_sections, minimized_content)
-                    VALUES (%s, %s, %s, %s, %s)
+                    INSERT INTO patents (patent_id, company_name, pdf_link, raw_sections, minimized_content, patent_tags)
+                    VALUES (%s, %s, %s, %s, %s, %s)
                     ON CONFLICT (patent_id) DO UPDATE SET
                         raw_sections = EXCLUDED.raw_sections,
-                        minimized_content = EXCLUDED.minimized_content
+                        minimized_content = EXCLUDED.minimized_content,
+                        patent_tags = EXCLUDED.patent_tags
                     """,
-                    (patent_id, company_name, pdf_link, json.dumps(raw_sections), minimized_content),
+                    (patent_id, company_name, pdf_link, json.dumps(raw_sections), minimized_content, patent_tags),
                 )
             conn.commit()
 
+    def update_patent_tags(self, patent_id: str, tags: List[str]) -> None:
+        """Update the technology domain tags for a patent."""
+        with self.get_conn() as conn:
+            with conn.cursor() as cursor:
+                cursor.execute(
+                    "UPDATE patents SET patent_tags = %s WHERE patent_id = %s",
+                    (tags, patent_id),
+                )
+            conn.commit()
+
+    def get_patents_by_tags(
+        self,
+        tags: List[str],
+        limit: int = 100,
+        offset: int = 0,
+    ) -> List[Dict]:
+        """Retrieve patents that match any of the given tags.
+
+        Args:
+            tags: List of tag strings to filter by (OR logic)
+            limit: Max results
+            offset: Pagination offset
+
+        Returns:
+            List of patent dicts
+        """
+        with self.get_conn() as conn:
+            with conn.cursor(cursor_factory=RealDictCursor) as cursor:
+                cursor.execute(
+                    """
+                    SELECT * FROM patents
+                    WHERE patent_tags && %s
+                    ORDER BY created_at DESC
+                    LIMIT %s OFFSET %s
+                    """,
+                    (tags, limit, offset),
+                )
+                return [dict(row) for row in cursor.fetchall()]
+
     def get_cached_serp_query(self, query_hash: str) -> Optional[List[str]]:
         """Look up cached SERP query results.
 

SPARC/llm.py

@@ -1,7 +1,8 @@
 """LLM integration for patent analysis using OpenRouter."""
 
+import json
 import logging
-from typing import Dict
+from typing import Dict, List
 
 from openai import OpenAI
 
@@ -245,4 +246,61 @@ Provide a comprehensive analysis (4-5 paragraphs) with a final verdict on the co
             metadata={**metadata, "pending": True}
         )
         return placeholder
-       
+
+    # Canonical technology domain tags
+    CANONICAL_TAGS = ["ai", "semiconductors", "materials", "biotech", "networking", "other"]
+
+    def classify_patent_tags(self, patent_content: str, model: str | None = None) -> List[str]:
+        """Classify a patent into one or more technology domain tags.
+
+        Sends the patent abstract/claims to the LLM with a classification prompt
+        and returns a list of canonical tags.
+
+        Args:
+          patent_content: Minimized patent text (abstract, claims, summary)
+          model: Optional model override
+
+        Returns:
+          List of canonical tag strings from CANONICAL_TAGS
+        """
+        prompt = f"""You are a patent classification system. Analyze the following patent content and assign one or more technology domain tags from this exact list:
+
+Tags: ai, semiconductors, materials, biotech, networking, other
+
+Rules:
+- Return ONLY a JSON array of tag strings, e.g. ["ai", "semiconductors"]
+- Use ONLY tags from the list above
+- Assign "other" only if no other tag fits
+- A patent can have multiple tags if it spans domains
+
+Patent Content:
+{patent_content}
+
+Return ONLY the JSON array, nothing else."""
+
+        effective_model = model or self.model
+
+        if self.test_mode:
+            logger.debug("TEST MODE - Classification prompt:\n%s", prompt)
+            return ["other"]
+
+        if self.client:
+            try:
+                response = self.client.chat.completions.create(
+                    model=effective_model,
+                    max_tokens=128,
+                    messages=[{"role": "user", "content": prompt}],
+                )
+                raw = response.choices[0].message.content.strip()
+                tags = json.loads(raw)
+                # Validate and filter to canonical tags only
+                valid_tags = [t for t in tags if t in self.CANONICAL_TAGS]
+                return valid_tags if valid_tags else ["other"]
+            except (json.JSONDecodeError, AttributeError, TypeError) as e:
+                logger.warning("Failed to parse classification response: %s", e)
+                return ["other"]
+            except Exception as e:
+                logger.warning("Classification LLM call failed: %s", e)
+                return ["other"]
+
+        return ["other"]

SPARC/types.py

@@ -25,6 +25,7 @@ class CompanyAnalysisResult:
     success: bool
     error: str | None = None
     model: str | None = None
+    tags: list[str] = field(default_factory=list)
     timestamp: datetime = field(default_factory=datetime.now)
 
 

frontend/src/api/client.ts

@@ -199,8 +199,18 @@ export const analyticsApi = {
     const response = await api.get<TrendData>(`/analytics/trends?days=${days}`);
     return response.data;
   },
+
+  getTagDistribution: async (): Promise<TagDistribution> => {
+    const response = await api.get<TagDistribution>('/analytics/tags');
+    return response.data;
+  },
 };
 
+export interface TagDistribution {
+  by_tag: Array<{ tag: string; count: number }>;
+  canonical_tags: string[];
+}
+
 // Admin API
 export const adminApi = {
   listUsers: async (limit = 100, offset = 0): Promise<User[]> => {

frontend/src/pages/Analytics.tsx

@@ -1,14 +1,24 @@
 import { useState } from 'react';
 import { useQuery } from '@tanstack/react-query';
 import { analyticsApi } from '../api/client';
-import { AlertCircle, Database } from 'lucide-react';
+import { AlertCircle, Database, Tag } from 'lucide-react';
 import { PieChart, Pie, Cell, BarChart, Bar, LineChart, Line, XAxis, YAxis, Tooltip, ResponsiveContainer, Legend } from 'recharts';
 import { useChartTheme } from '../context/useChartTheme';
 
 const COLORS = ['#6366f1', '#0ea5e9', '#10b981', '#f59e0b', '#ef4444', '#8b5cf6', '#ec4899', '#14b8a6'];
 
+const TAG_COLORS: Record<string, string> = {
+  ai: '#6366f1',
+  semiconductors: '#0ea5e9',
+  materials: '#10b981',
+  biotech: '#f59e0b',
+  networking: '#ec4899',
+  other: '#8b5cf6',
+};
+
 export function AnalyticsPage() {
   const [days, setDays] = useState(30);
+  const [selectedTags, setSelectedTags] = useState<string[]>([]);
   const chartTheme = useChartTheme();
 
   const { data, isLoading, isError, refetch } = useQuery({
@@ -21,6 +31,17 @@ export function AnalyticsPage() {
     queryFn: () => analyticsApi.getTrends(days),
   });
 
+  const tagQuery = useQuery({
+    queryKey: ['analytics-tags'],
+    queryFn: () => analyticsApi.getTagDistribution(),
+  });
+
+  const toggleTag = (tag: string) => {
+    setSelectedTags((prev) =>
+      prev.includes(tag) ? prev.filter((t) => t !== tag) : [...prev, tag]
+    );
+  };
+
   if (isLoading) {
     return (
       <div className="space-y-6">
@@ -107,6 +128,13 @@ export function AnalyticsPage() {
     count: t.count,
   }));
 
+  const tagData = tagQuery.data?.by_tag?.map((t) => ({
+    name: t.tag,
+    count: t.count,
+  })) || [];
+
+  const canonicalTags = tagQuery.data?.canonical_tags || [];
+
   return (
     <div className="space-y-6">
       {/* Header */}
@@ -131,6 +159,50 @@ export function AnalyticsPage() {
         </select>
       </div>
 
+      {/* Tag Filter Controls */}
+      {canonicalTags.length > 0 && (
+        <div className="bg-bg-card/60 border border-primary/15 rounded-2xl p-4">
+          <div className="flex items-center gap-2 mb-3">
+            <Tag size={16} className="text-primary" />
+            <span className="text-sm font-semibold text-text-primary">Filter by Technology Domain</span>
+            {selectedTags.length > 0 && (
+              <button
+                onClick={() => setSelectedTags([])}
+                className="ml-auto text-xs text-text-secondary hover:text-primary transition-colors"
+              >
+                Clear all
+              </button>
+            )}
+          </div>
+          <div className="flex flex-wrap gap-2">
+            {canonicalTags.map((tag) => {
+              const isActive = selectedTags.includes(tag);
+              const color = TAG_COLORS[tag] || '#8b5cf6';
+              const tagCount = tagQuery.data?.by_tag?.find((t) => t.tag === tag)?.count || 0;
+              return (
+                <button
+                  key={tag}
+                  onClick={() => toggleTag(tag)}
+                  className={`px-3 py-1.5 rounded-lg text-sm font-medium transition-all ${
+                    isActive
+                      ? 'text-white shadow-md'
+                      : 'bg-bg-card/80 text-text-secondary border border-primary/20 hover:border-primary/40'
+                  }`}
+                  style={isActive ? { backgroundColor: color } : {}}
+                >
+                  {tag}
+                  {tagCount > 0 && (
+                    <span className={`ml-1.5 text-xs ${isActive ? 'opacity-80' : 'opacity-60'}`}>
+                      ({tagCount})
+                    </span>
+                  )}
+                </button>
+              );
+            })}
+          </div>
+        </div>
+      )}
+
       {/* Summary Metrics */}
       <div className="grid grid-cols-1 md:grid-cols-3 gap-4">
         <MetricCard label="Total Analyses" value={data.total_messages} />
@@ -187,6 +259,28 @@ export function AnalyticsPage() {
             </ResponsiveContainer>
           </div>
         )}
+
+        {/* Bar Chart - Technology Domain Tags */}
+        {tagData.length > 0 && (
+          <div className="bg-bg-card/60 border border-primary/15 rounded-2xl p-6">
+            <h3 className="text-lg font-semibold text-text-primary mb-4">Patents by Technology Domain</h3>
+            <ResponsiveContainer width="100%" height={300}>
+              <BarChart data={tagData}>
+                <XAxis dataKey="name" stroke={chartTheme.axisStroke} fontSize={12} />
+                <YAxis stroke={chartTheme.axisStroke} fontSize={12} />
+                <Tooltip
+                  contentStyle={chartTheme.tooltipContentStyle}
+                  labelStyle={chartTheme.tooltipLabelStyle}
+                />
+                <Bar dataKey="count" radius={[4, 4, 0, 0]}>
+                  {tagData.map((entry, index) => (
+                    <Cell key={`tag-cell-${index}`} fill={TAG_COLORS[entry.name] || COLORS[index % COLORS.length]} />
+                  ))}
+                </Bar>
+              </BarChart>
+            </ResponsiveContainer>
+          </div>
+        )}
       </div>
 
       {/* Trend Charts */}

tests/test_patent_tags.py

@@ -0,0 +1,262 @@
+"""Tests for LLM-based patent classification tagging by technology domain."""
+
+import json
+from unittest.mock import MagicMock, Mock, patch
+
+import pytest
+
+from SPARC.llm import LLMAnalyzer
+
+
+class TestClassifyPatentTags:
+    """Test the classify_patent_tags method on LLMAnalyzer."""
+
+    @pytest.fixture(autouse=True)
+    def mock_database(self, mocker):
+        """Mock the database client for all tests."""
+        mock_db_client = Mock()
+        mock_db_client.get_cached_response.return_value = None
+        mock_db_client.store_message.return_value = 1
+        mocker.patch("SPARC.llm.DatabaseClient", return_value=mock_db_client)
+        return mock_db_client
+
+    def test_classify_returns_valid_tags(self, mocker, mock_database):
+        """Test that classify_patent_tags returns valid canonical tags from LLM."""
+        mock_openai = mocker.patch("SPARC.llm.OpenAI")
+        mock_client = Mock()
+        mock_openai.return_value = mock_client
+
+        mock_response = Mock()
+        mock_response.choices = [Mock(message=Mock(content='["ai", "semiconductors"]'))]
+        mock_client.chat.completions.create.return_value = mock_response
+
+        analyzer = LLMAnalyzer(api_key="test-key")
+        tags = analyzer.classify_patent_tags("A patent about neural network accelerator chips")
+
+        assert tags == ["ai", "semiconductors"]
+        mock_client.chat.completions.create.assert_called_once()
+
+        # Verify the prompt contains classification instructions
+        call_args = mock_client.chat.completions.create.call_args
+        prompt_text = call_args[1]["messages"][0]["content"]
+        assert "ai" in prompt_text
+        assert "semiconductors" in prompt_text
+        assert "JSON array" in prompt_text
+
+    def test_classify_filters_invalid_tags(self, mocker, mock_database):
+        """Test that invalid tags from LLM response are filtered out."""
+        mock_openai = mocker.patch("SPARC.llm.OpenAI")
+        mock_client = Mock()
+        mock_openai.return_value = mock_client
+
+        mock_response = Mock()
+        mock_response.choices = [Mock(message=Mock(content='["ai", "quantum_computing", "biotech"]'))]
+        mock_client.chat.completions.create.return_value = mock_response
+
+        analyzer = LLMAnalyzer(api_key="test-key")
+        tags = analyzer.classify_patent_tags("Some patent content")
+
+        assert tags == ["ai", "biotech"]
+        assert "quantum_computing" not in tags
+
+    def test_classify_returns_other_when_all_invalid(self, mocker, mock_database):
+        """Test fallback to 'other' when all LLM tags are invalid."""
+        mock_openai = mocker.patch("SPARC.llm.OpenAI")
+        mock_client = Mock()
+        mock_openai.return_value = mock_client
+
+        mock_response = Mock()
+        mock_response.choices = [Mock(message=Mock(content='["quantum", "robotics"]'))]
+        mock_client.chat.completions.create.return_value = mock_response
+
+        analyzer = LLMAnalyzer(api_key="test-key")
+        tags = analyzer.classify_patent_tags("Some patent content")
+
+        assert tags == ["other"]
+
+    def test_classify_handles_malformed_json(self, mocker, mock_database):
+        """Test graceful handling of non-JSON LLM response."""
+        mock_openai = mocker.patch("SPARC.llm.OpenAI")
+        mock_client = Mock()
+        mock_openai.return_value = mock_client
+
+        mock_response = Mock()
+        mock_response.choices = [Mock(message=Mock(content="ai, semiconductors"))]
+        mock_client.chat.completions.create.return_value = mock_response
+
+        analyzer = LLMAnalyzer(api_key="test-key")
+        tags = analyzer.classify_patent_tags("Some patent content")
+
+        assert tags == ["other"]
+
+    def test_classify_handles_api_error(self, mocker, mock_database):
+        """Test graceful fallback when LLM API call fails."""
+        mock_openai = mocker.patch("SPARC.llm.OpenAI")
+        mock_client = Mock()
+        mock_openai.return_value = mock_client
+
+        mock_client.chat.completions.create.side_effect = Exception("API timeout")
+
+        analyzer = LLMAnalyzer(api_key="test-key")
+        tags = analyzer.classify_patent_tags("Some patent content")
+
+        assert tags == ["other"]
+
+    def test_classify_test_mode(self, mocker, mock_database):
+        """Test that test mode returns 'other' without API call."""
+        mocker.patch("SPARC.llm.config")
+
+        analyzer = LLMAnalyzer(test_mode=True)
+        tags = analyzer.classify_patent_tags("Some patent content")
+
+        assert tags == ["other"]
+
+    def test_classify_no_api_client(self, mocker, mock_database):
+        """Test that without API client, classification returns 'other'."""
+        mocker.patch("SPARC.llm.config")
+
+        analyzer = LLMAnalyzer(use_cache=False)
+        tags = analyzer.classify_patent_tags("Some patent content")
+
+        assert tags == ["other"]
+
+    def test_canonical_tags_list(self):
+        """Test that the canonical tag list matches requirements."""
+        expected = ["ai", "semiconductors", "materials", "biotech", "networking", "other"]
+        assert LLMAnalyzer.CANONICAL_TAGS == expected
+
+    def test_classify_uses_model_override(self, mocker, mock_database):
+        """Test that model override is passed to the API call."""
+        mock_openai = mocker.patch("SPARC.llm.OpenAI")
+        mock_client = Mock()
+        mock_openai.return_value = mock_client
+
+        mock_response = Mock()
+        mock_response.choices = [Mock(message=Mock(content='["ai"]'))]
+        mock_client.chat.completions.create.return_value = mock_response
+
+        analyzer = LLMAnalyzer(api_key="test-key")
+        analyzer.classify_patent_tags("content", model="openai/gpt-4o")
+
+        call_args = mock_client.chat.completions.create.call_args
+        assert call_args[1]["model"] == "openai/gpt-4o"
+
+
+class TestPatentTagsStorage:
+    """Test that tags are persisted to the database."""
+
+    @pytest.fixture(autouse=True)
+    def mock_db(self, mocker):
+        """Mock DatabaseClient for all tests."""
+        mock_db_cls = mocker.patch("SPARC.analyzer.DatabaseClient")
+        mock_db_instance = MagicMock()
+        mock_db_instance.get_cached_patent.return_value = None
+        mock_db_instance.get_cached_serp_query.return_value = None
+        mock_db_cls.return_value = mock_db_instance
+        return mock_db_instance
+
+    def test_tags_stored_after_classification(self, mocker, mock_db):
+        """Test that classify_patent_tags results are stored via update_patent_tags."""
+        from SPARC.analyzer import CompanyAnalyzer
+        from SPARC.types import Patent, Patents
+
+        mock_query = mocker.patch("SPARC.analyzer.SERP.query")
+        mock_save = mocker.patch("SPARC.analyzer.SERP.save_patents")
+        mock_parse = mocker.patch("SPARC.analyzer.SERP.parse_patent_pdf")
+        mock_minimize = mocker.patch("SPARC.analyzer.SERP.minimize_patent_for_llm")
+        mock_llm_cls = mocker.patch("SPARC.analyzer.LLMAnalyzer")
+
+        patent = Patent(patent_id="US123", pdf_link="http://example.com/test.pdf")
+        mock_query.return_value = Patents(patents=[patent])
+
+        def save_side_effect(p):
+            p.pdf_path = "patents/US123.pdf"
+            return p
+
+        mock_save.side_effect = save_side_effect
+        mock_parse.return_value = {"abstract": "Test abstract"}
+        mock_minimize.return_value = "Minimized content"
+
+        mock_llm_instance = Mock()
+        mock_llm_instance.analyze_patent_portfolio.return_value = "Analysis result"
+        mock_llm_instance.classify_patent_tags.return_value = ["ai", "semiconductors"]
+        mock_llm_cls.return_value = mock_llm_instance
+
+        analyzer = CompanyAnalyzer()
+        analyzer.analyze_company("TestCorp")
+
+        # Verify classification was called
+        mock_llm_instance.classify_patent_tags.assert_called_once_with(
+            patent_content="Minimized content", model=None
+        )
+
+        # Verify tags were persisted
+        mock_db.update_patent_tags.assert_called_once_with("US123", ["ai", "semiconductors"])
+
+    def test_cached_tags_skip_classification(self, mocker, mock_db):
+        """Test that patents with cached tags skip re-classification."""
+        from SPARC.analyzer import CompanyAnalyzer
+        from SPARC.types import Patent, Patents
+
+        mocker.patch("SPARC.analyzer.SERP.query")
+        mocker.patch("SPARC.analyzer.SERP.save_patents")
+        mock_llm_cls = mocker.patch("SPARC.analyzer.LLMAnalyzer")
+
+        # Simulate DB cache hit with tags
+        mock_db.get_cached_patent.return_value = {
+            "patent_id": "US123",
+            "minimized_content": "Cached content",
+            "patent_tags": ["biotech"],
+        }
+        mock_db.get_cached_serp_query.return_value = ["US123"]
+
+        mock_llm_instance = Mock()
+        mock_llm_instance.analyze_patent_portfolio.return_value = "Analysis"
+        mock_llm_cls.return_value = mock_llm_instance
+
+        analyzer = CompanyAnalyzer()
+        analyzer.analyze_company("TestCorp")
+
+        # Should NOT classify since tags already present from cache
+        mock_llm_instance.classify_patent_tags.assert_not_called()
+        mock_db.update_patent_tags.assert_not_called()
+
+    def test_tags_in_analysis_result(self, mocker, mock_db):
+        """Test that tags appear in the CompanyAnalysisResult."""
+        from SPARC.analyzer import CompanyAnalyzer
+        from SPARC.types import Patent, Patents
+
+        mock_query = mocker.patch("SPARC.analyzer.SERP.query")
+        mock_save = mocker.patch("SPARC.analyzer.SERP.save_patents")
+        mock_parse = mocker.patch("SPARC.analyzer.SERP.parse_patent_pdf")
+        mock_minimize = mocker.patch("SPARC.analyzer.SERP.minimize_patent_for_llm")
+        mock_llm_cls = mocker.patch("SPARC.analyzer.LLMAnalyzer")
+
+        patent = Patent(patent_id="US123", pdf_link="http://example.com/test.pdf")
+        mock_query.return_value = Patents(patents=[patent])
+
+        def save_side_effect(p):
+            p.pdf_path = "patents/US123.pdf"
+            return p
+
+        mock_save.side_effect = save_side_effect
+        mock_parse.return_value = {"abstract": "Test abstract"}
+        mock_minimize.return_value = "Minimized content"
+
+        mock_llm_instance = Mock()
+        mock_llm_instance.analyze_patent_portfolio.return_value = "Strong innovation"
+        mock_llm_instance.classify_patent_tags.return_value = ["ai", "networking"]
+        mock_llm_cls.return_value = mock_llm_instance
+
+        # After analysis, get_cached_patent returns the patent with tags
+        mock_db.get_cached_serp_query.side_effect = [None, ["US123"]]
+        mock_db.get_cached_patent.side_effect = [
+            None,  # First call during _process_single_patent
+            {"patent_id": "US123", "patent_tags": ["ai", "networking"]},  # Second call during _analyze_company_safe
+        ]
+
+        analyzer = CompanyAnalyzer()
+        result = analyzer._analyze_company_safe("TestCorp")
+
+        assert result.success is True
+        assert result.tags == ["ai", "networking"]