fix(ai): increase timeout to 60s, keep model resident in RAM for 24h, and warm up on startup

app/ai.py

@@ -223,6 +223,7 @@ async def translate_sentence_ai(
         "prompt": prompt,
         "stream": False,
         "format": "json",
+        "keep_alive": "24h",
         "options": {
             "temperature": 0.2,
             "num_predict": 300,
@@ -234,7 +235,7 @@ async def translate_sentence_ai(
     source = f"ollama:{config.OLLAMA_MODEL}"
 
     try:
-        async with httpx.AsyncClient(timeout=35.0) as client:
+        async with httpx.AsyncClient(timeout=60.0) as client:
             resp = await client.post(f"{config.OLLAMA_BASE_URL}/api/generate", json=payload)
             if resp.status_code == 200:
                 data = resp.json()
@@ -282,7 +283,8 @@ async def translate_sentence_ai(
             else:
                 logger.warning(f"Ollama returned HTTP {resp.status_code}: {resp.text}")
     except Exception as e:
-        logger.warning(f"Ollama connection error: {e}")
+        err_msg = f"{type(e).__name__}: {e}" if str(e) else type(e).__name__
+        logger.warning(f"Ollama connection error ({err_msg})")
 
     # 4. Graceful Fallback if Ollama is unreachable
     if not sentence_ru:
@@ -329,3 +331,22 @@ async def get_ollama_status() -> Dict[str, Any]:
         "models": [],
         "active_model": config.OLLAMA_MODEL,
     }
+
+
+async def warmup_ollama_model() -> bool:
+    """Preloads the configured Ollama model into RAM with keep_alive to eliminate cold-start latency."""
+    try:
+        async with httpx.AsyncClient(timeout=60.0) as client:
+            resp = await client.post(
+                f"{config.OLLAMA_BASE_URL}/api/generate",
+                json={"model": config.OLLAMA_MODEL, "keep_alive": "24h"},
+            )
+            if resp.status_code == 200:
+                logger.info(f"Ollama model '{config.OLLAMA_MODEL}' warmed up successfully (keep_alive=24h).")
+                return True
+            else:
+                logger.warning(f"Ollama warmup returned HTTP {resp.status_code}: {resp.text}")
+    except Exception as e:
+        err_msg = f"{type(e).__name__}: {e}" if str(e) else type(e).__name__
+        logger.warning(f"Ollama warmup deferred ({err_msg})")
+    return False

app/main.py

@@ -35,7 +35,7 @@ from app.reader import (
     fetch_and_parse_rfc, parse_rfc_text, fetch_and_parse_wikipedia,
     extract_wikipedia_title_and_lang
 )
-from app.ai import translate_sentence_ai, get_ollama_status
+from app.ai import translate_sentence_ai, get_ollama_status, warmup_ollama_model
 from app.phrasal import detect_phrasal_verb
 from app.cefr import (
     get_cefr_level_summaries,
@@ -68,6 +68,7 @@ from app.miner import run_mining_task_background
 async def lifespan(app: FastAPI):
     init_db()
     ensure_admin_user()
+    asyncio.create_task(warmup_ollama_model())
     yield
     await close_http_client()
 

docker-compose.yml

@@ -29,4 +29,4 @@ services:
     volumes:
       - ./data/ollama:/root/.ollama
     environment:
-      - OLLAMA_KEEP_ALIVE=5m
+      - OLLAMA_KEEP_ALIVE=24h