feat: configurable LLM model, SERP cache TTL, structured logging, fix patent_id type

- Make LLM model configurable via MODEL env var, default anthropic/claude-3.5-sonnet (#12)
- Expose SERP cache TTL as SERP_CACHE_TTL_HOURS env var, default 24 hours (#13)
- Fix Patent.patent_id type annotation from int to str in types.py (#14)
- Replace all print() calls with structured logging in analyzer.py and llm.py (#11)
- Add LOG_LEVEL config with basicConfig setup in config.py
- Add model and serp_cache_ttl_hours to config.py

Closes leeworks-agents/SPARC#11
Closes leeworks-agents/SPARC#12
Closes leeworks-agents/SPARC#13
Closes leeworks-agents/SPARC#14

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

ROADMAP.md

@@ -0,0 +1,122 @@
+# SPARC Roadmap
+
+Semiconductor Patent & Analytics Report Core -- development priorities.
+
+## Current State
+
+SPARC is a patent analysis platform with a working end-to-end pipeline:
+Python/FastAPI backend, React/TypeScript frontend, PostgreSQL for persistence
+and caching, Docker Compose for local development, and Gitea Actions CI/CD for
+image builds. Core features (patent retrieval via SerpAPI, PDF parsing, LLM
+analysis via OpenRouter/Claude, batch processing, JWT authentication, analytics
+dashboard) are all implemented and functional.
+
+---
+
+## P1 -- High Priority
+
+These items address correctness, security, and reliability gaps that should be
+resolved before broader production use.
+
+### Security hardening
+
+- **Rotate default JWT secret.** `auth.py` ships a fallback
+  `sparc-secret-key-change-in-production` that will be used if `JWT_SECRET` is
+  unset. Add a startup check that refuses to start with the default secret in
+  non-development environments.
+- **CORS allow-origins are hardcoded.** `api.py` only permits
+  `localhost:3000` and `localhost:5173`. Make the allowed origins configurable
+  via environment variable so the dashboard works when deployed behind a real
+  domain.
+- **Database credentials in docker-compose.yml.** The compose file embeds
+  `postgres:postgres` in plain text. Reference a `.env` file or Docker secrets
+  instead.
+
+### Error handling and resilience
+
+- **`get_db_client()` in `auth.py` creates a new `DatabaseClient` on every
+  call.** This bypasses the connection pool and can exhaust database
+  connections under load. Refactor to share a single pooled client.
+- **`_jobs` dict is in-memory only.** Job state is lost on API restart. Persist
+  job status in PostgreSQL or Redis so async batch results survive restarts.
+- **No rate limiting on auth endpoints.** `/auth/login` and `/auth/register`
+  are unprotected against brute-force or abuse. Add rate limiting middleware.
+
+### Test coverage for auth and admin
+
+- The existing API tests (`tests/test_api.py`) bypass authentication entirely.
+  Add tests that exercise the JWT flow: registration, login, protected-route
+  access, token refresh, and admin-only endpoints.
+
+---
+
+## P2 -- Medium Priority
+
+Improvements to usability, performance, and developer experience.
+
+### Backend
+
+- **Add structured logging.** Replace `print()` calls throughout `analyzer.py`,
+  `serp_api.py`, and `llm.py` with Python `logging` so log levels and
+  formatting are consistent.
+- **Make LLM model configurable.** `llm.py` hardcodes
+  `anthropic/claude-3.5-sonnet`. Accept a `MODEL` environment variable to allow
+  switching models without code changes.
+- **SERP cache TTL is hardcoded to 24 hours.** Expose `SERP_CACHE_TTL_HOURS`
+  as an environment variable in `config.py`.
+- **Patent PDF storage.** PDFs are saved to a local `patents/` directory. For
+  containerized deployments, consider object storage (S3/MinIO) or at minimum
+  document the volume mount requirement more prominently.
+- **`analyze_single_patent` assumes local file path.** The method constructs
+  `patents/{patent_id}.pdf` and reads from disk, but does not download the PDF
+  first. Either integrate the download step or document the prerequisite.
+- **`Patent.patent_id` typed as `int` in `types.py` but used as `str`
+  everywhere.** Fix the type annotation to `str`.
+
+### Frontend
+
+- **No loading/error states on several pages.** The Batch and Analytics pages
+  would benefit from skeleton loaders and user-friendly error messages.
+- **No dark mode.** Tailwind is configured but no dark variant is applied.
+- **Missing `package-lock.json` or `pnpm-lock.yaml`.** The frontend has no
+  lockfile committed, leading to non-reproducible builds.
+
+### CI/CD
+
+- **No test stage in the Gitea Actions workflow.** `build.yaml` builds and
+  pushes images but never runs `pytest`. Add a test job that gates the build.
+- **No linting or type checking.** Add `ruff` (Python) and `tsc --noEmit`
+  (TypeScript) to CI.
+
+---
+
+## P3 -- Nice to Have
+
+Lower-urgency enhancements and future features.
+
+- **Export analysis reports.** Allow users to download analysis results as PDF
+  or CSV from the dashboard.
+- **Comparison view.** Side-by-side comparison of two companies' patent
+  portfolios.
+- **Scheduled/recurring analysis.** Periodically re-analyze tracked companies
+  and alert on significant changes.
+- **Webhook/notification support.** Send alerts (Slack, Discord, email) when
+  batch jobs complete or when a company's innovation score changes
+  significantly.
+- **Multi-model support.** Let users choose between LLM providers per analysis
+  (e.g., GPT-4o, Gemini, Claude) and compare outputs.
+- **Patent trend charts.** Visualize patent filing frequency and technology
+  category distribution over time in the Analytics page.
+- **API pagination.** The `/analyze/batch` and `/jobs` endpoints could benefit
+  from cursor-based pagination for large result sets.
+- **OpenAPI client generation.** Auto-generate the TypeScript API client from
+  the FastAPI OpenAPI spec to keep frontend types in sync.
+
+---
+
+## Infrastructure and Deployment
+
+Kubernetes manifests, Helm charts, and cluster-level concerns (MetalLB,
+storage, FluxCD sync) are tracked in the
+[Talos](https://10.0.1.10/leeworks-agents/Talos) repository. File
+infrastructure-related issues there, not here.

SPARC/analyzer.py

@@ -5,10 +5,13 @@ to provide company performance estimation based on patent portfolios.
 """
 
 import hashlib
+import logging
 from concurrent.futures import ThreadPoolExecutor, as_completed
 from typing import Callable
 
 from SPARC import config
+
+logger = logging.getLogger(__name__)
 from SPARC.database import DatabaseClient
 from SPARC.serp_api import SERP
 from SPARC.llm import LLMAnalyzer
@@ -52,13 +55,13 @@ class CompanyAnalyzer:
             query_hash = hashlib.sha256(company_name.lower().encode()).hexdigest()
             cached_ids = self.db.get_cached_serp_query(query_hash)
             if cached_ids is not None:
-                print(f"Using cached SERP results for {company_name} ({len(cached_ids)} patents)")
+                logger.info("Using cached SERP results for %s (%d patents)", company_name, len(cached_ids))
                 patents = Patents(patents=[
                     Patent(patent_id=pid, pdf_link="")
                     for pid in cached_ids
                 ])
             else:
-                print(f"Retrieving patents for {company_name}...")
+                logger.info("Retrieving patents for %s...", company_name)
                 patents = SERP.query(company_name)
                 # Cache the SERP results
                 if patents.patents:
@@ -66,12 +69,13 @@ class CompanyAnalyzer:
                         company_name=company_name,
                         query_hash=query_hash,
                         patent_ids=[p.patent_id for p in patents.patents],
+                        ttl_hours=config.serp_cache_ttl_hours,
                     )
 
         if not patents.patents:
             return f"No patents found for {company_name}"
 
-        print(f"Found {len(patents.patents)} patents. Processing...")
+        logger.info("Found %d patents. Processing...", len(patents.patents))
 
         # Download, parse, and minimize patents in parallel
         processed_patents = []
@@ -87,12 +91,12 @@ class CompanyAnalyzer:
                     if result:
                         processed_patents.append(result)
                 except Exception as e:
-                    print(f"Warning: Failed to process {patent.patent_id}: {e}")
+                    logger.warning("Failed to process %s: %s", patent.patent_id, e)
 
         if not processed_patents:
             return f"Failed to process any patents for {company_name}"
 
-        print(f"Analyzing portfolio with LLM...")
+        logger.info("Analyzing portfolio with LLM...")
 
         # Analyze the full portfolio with LLM
         analysis = self.llm_analyzer.analyze_patent_portfolio(
@@ -115,7 +119,7 @@ class CompanyAnalyzer:
         """
         # Note: This simplified version assumes the patent PDF is already downloaded
         # A more complete implementation would support direct patent ID lookup
-        print(f"Analyzing patent {patent_id} for {company_name}...")
+        logger.info("Analyzing patent %s for %s...", patent_id, company_name)
 
         patent_path = f"patents/{patent_id}.pdf"
 
@@ -169,7 +173,7 @@ class CompanyAnalyzer:
 
             return {"patent_id": patent.patent_id, "content": minimized_content}
         except Exception as e:
-            print(f"Warning: Failed to process {patent.patent_id}: {e}")
+            logger.warning("Failed to process %s: %s", patent.patent_id, e)
             return None
 
     def _analyze_company_safe(self, company_name: str) -> CompanyAnalysisResult:
@@ -240,7 +244,7 @@ class CompanyAnalyzer:
         results: list[CompanyAnalysisResult] = []
         total = len(companies)
 
-        print(f"Starting batch analysis of {total} companies...")
+        logger.info("Starting batch analysis of %d companies...", total)
 
         with ThreadPoolExecutor(max_workers=max_workers) as executor:
             future_to_company = {
@@ -257,8 +261,8 @@ class CompanyAnalyzer:
                     result = future.result()
                     results.append(result)
 
-                    status = "✓" if result.success else "✗"
-                    print(f"[{completed}/{total}] {status} {company}")
+                    status = "OK" if result.success else "FAIL"
+                    logger.info("[%d/%d] %s %s", completed, total, status, company)
 
                     if progress_callback:
                         progress_callback(company, completed, total)
@@ -273,12 +277,12 @@ class CompanyAnalyzer:
                             error=str(e),
                         )
                     )
-                    print(f"[{completed}/{total}] ✗ {company}: {e}")
+                    logger.error("[%d/%d] FAIL %s: %s", completed, total, company, e)
 
         successful = sum(1 for r in results if r.success)
         failed = total - successful
 
-        print(f"\nBatch complete: {successful} succeeded, {failed} failed")
+        logger.info("Batch complete: %d succeeded, %d failed", successful, failed)
 
         return BatchAnalysisResult(
             results=results,
@@ -304,20 +308,20 @@ class CompanyAnalyzer:
         results: list[CompanyAnalysisResult] = []
         total = len(companies)
 
-        print(f"Starting sequential analysis of {total} companies...")
+        logger.info("Starting sequential analysis of %d companies...", total)
 
         for idx, company in enumerate(companies, 1):
-            print(f"\n[{idx}/{total}] Analyzing {company}...")
+            logger.info("[%d/%d] Analyzing %s...", idx, total, company)
             result = self._analyze_company_safe(company)
             results.append(result)
 
-            status = "✓" if result.success else "✗"
-            print(f"[{idx}/{total}] {status} {company}")
+            status = "OK" if result.success else "FAIL"
+            logger.info("[%d/%d] %s %s", idx, total, status, company)
 
         successful = sum(1 for r in results if r.success)
         failed = total - successful
 
-        print(f"\nBatch complete: {successful} succeeded, {failed} failed")
+        logger.info("Batch complete: %d succeeded, %d failed", successful, failed)
 
         return BatchAnalysisResult(
             results=results,

SPARC/config.py

@@ -2,11 +2,20 @@
 
 Loads environment variables from .env file for API keys and other secrets.
 """
-from dotenv import load_dotenv
+import logging
 import os
 
+from dotenv import load_dotenv
+
 load_dotenv()
 
+# Logging configuration
+log_level = os.getenv("LOG_LEVEL", "INFO").upper()
+logging.basicConfig(
+    level=getattr(logging, log_level, logging.INFO),
+    format="%(asctime)s %(levelname)s %(name)s %(message)s",
+)
+
 # SerpAPI key for patent search
 api_key = os.getenv("API_KEY")
 
@@ -30,6 +39,12 @@ use_database = os.getenv("USE_DATABASE", "false").lower() in ("true", "1", "yes"
 patent_search_days = int(os.getenv("PATENT_SEARCH_DAYS", "90"))
 patent_thread_workers = int(os.getenv("PATENT_THREAD_WORKERS", "5"))
 
+# LLM model to use via OpenRouter (e.g. "anthropic/claude-3.5-sonnet", "openai/gpt-4o")
+model = os.getenv("MODEL", "anthropic/claude-3.5-sonnet")
+
+# SERP cache TTL in hours (how long cached search results are considered fresh)
+serp_cache_ttl_hours = int(os.getenv("SERP_CACHE_TTL_HOURS", "24"))
+
 # Root path for running behind a reverse proxy (e.g., "/api" when served at /api/)
 # This ensures OpenAPI docs work correctly when accessed via the proxy
 root_path = os.getenv("ROOT_PATH", "")

SPARC/llm.py

@@ -1,9 +1,14 @@
 """LLM integration for patent analysis using OpenRouter."""
 
+import logging
+from typing import Dict
+
 from openai import OpenAI
+
 from SPARC import config
 from SPARC.database import DatabaseClient
-from typing import Dict
+
+logger = logging.getLogger(__name__)
 
 
 class LLMAnalyzer:
@@ -20,7 +25,7 @@ class LLMAnalyzer:
         """
         self.test_mode = test_mode
         self.use_cache = use_cache if use_cache is not None else config.use_cache
-        self.model = "anthropic/claude-3.5-sonnet"
+        self.model = config.model
 
         # Always initialize database client for storage and caching
         self.db_client = DatabaseClient(config.database_url)
@@ -59,11 +64,7 @@ Patent Content:
 Provide a concise analysis (2-3 paragraphs) focusing on what this patent reveals about the company's technical direction and competitive advantage."""
 
         if self.test_mode:
-            print("=" * 80)
-            print("TEST MODE - Prompt that would be sent to LLM:")
-            print("=" * 80)
-            print(prompt)
-            print("=" * 80)
+            logger.debug("TEST MODE - Prompt that would be sent to LLM:\n%s", prompt)
             return "[TEST MODE - No API call made]"
 
         # Check cache first
@@ -165,7 +166,7 @@ Patent Portfolio:
 Provide a comprehensive analysis (4-5 paragraphs) with a final verdict on the company's innovation strength and performance outlook."""
 
         if self.test_mode:
-            print(prompt)
+            logger.debug("TEST MODE - Portfolio prompt:\n%s", prompt)
             return "[TEST MODE]"
 
         metadata = {

SPARC/types.py

@@ -4,7 +4,7 @@ from datetime import datetime
 
 @dataclass
 class Patent:
-    patent_id: int
+    patent_id: str
     pdf_link: str
     pdf_path: str | None = None
     summary: dict | None = None