fix: punctuate raw transcript, strip JSON code fences, filter null speaker names

- llm: punctuate() adds punctuation/capitalisation without changing words
- llm: _strip_code_fences() handles markdown-wrapped JSON from gemma3
- llm: filter string 'null' from identify_speakers result
- pipeline: punctuate raw_text in parallel with refine for solo recordings

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

api/pipeline.py

@@ -74,15 +74,14 @@ async def _run_solo_pipeline(cfg, wav_path, output_dir, instructions):
     await broadcast({"event": "transcribed", "raw": raw_text})
 
     client = OllamaClient(base_url=cfg["ollama"]["base_url"])
-    refined = await client.refine(
+    punctuated, refined = await asyncio.gather(
-        raw_text=raw_text,
+        client.punctuate(raw_text, model=cfg["ollama"]["model"]),
-        instructions=instructions,
+        client.refine(raw_text=raw_text, instructions=instructions, model=cfg["ollama"]["model"]),
-        model=cfg["ollama"]["model"],
     )
+    title, tldr = await client.generate_title_and_tldr(refined, model=cfg["ollama"]["model"])
 
     dt = datetime.now()
-    title, tldr = await client.generate_title_and_tldr(refined, model=cfg["ollama"]["model"])
+    paths = write_solo_docs(raw_text=punctuated, refined=refined, output_dir=output_dir, dt=dt, title=title, tldr=tldr)
-    paths = write_solo_docs(raw_text=raw_text, refined=refined, output_dir=output_dir, dt=dt, title=title, tldr=tldr)
 
     await state.set_status(Status.IDLE)
     await broadcast({

llm.py

@@ -46,6 +46,21 @@ Zweiter Absatz...
 """
 
 
+PUNCTUATE_PROMPT = """Du bekommst einen rohen deutschen Sprachtranskript ohne Großschreibung und Satzzeichen.
+Füge AUSSCHLIESSLICH Satzzeichen (Punkt, Komma, Fragezeichen, Ausrufezeichen) und Großschreibung am Satzanfang hinzu.
+Verändere KEINE Wörter, kürze NICHTS, füge NICHTS hinzu.
+Antworte NUR mit dem korrigierten Text, ohne Kommentar."""
+
+
+def _strip_code_fences(text: str) -> str:
+    """Remove markdown code fences (```json ... ```) from LLM responses."""
+    import re
+    m = re.search(r"```(?:json)?\s*([\s\S]*?)\s*```", text)
+    if m:
+        return m.group(1)
+    return text
+
+
 class OllamaClient:
     def __init__(self, base_url: str = "http://localhost:11434"):
         self.base_url = base_url
@@ -91,7 +106,7 @@ class OllamaClient:
                 },
             )
             r.raise_for_status()
-            raw = r.json()["response"].strip()
+            raw = _strip_code_fences(r.json()["response"].strip())
         try:
             data = json.loads(raw)
             title = str(data.get("title", "")).strip() or "Diktat"
@@ -100,6 +115,21 @@ class OllamaClient:
         except Exception:
             return "Diktat", "Kein TL;DR verfügbar."
 
+    async def punctuate(
+        self,
+        text: str,
+        model: str = "gemma3:12b",
+    ) -> str:
+        """Add punctuation and capitalisation to raw whisper output without changing words."""
+        async with httpx.AsyncClient(timeout=120) as client:
+            r = await client.post(
+                f"{self.base_url}/api/generate",
+                json={"model": model, "prompt": text, "system": PUNCTUATE_PROMPT, "stream": False},
+            )
+            r.raise_for_status()
+            result = r.json()["response"].strip()
+        return result if result else text
+
     async def identify_speakers(
         self,
         transcript_excerpt: str,
@@ -123,7 +153,7 @@ class OllamaClient:
             data = json.loads(raw)
             if not isinstance(data, dict):
                 return {}
-            return {k: v for k, v in data.items() if v}
+            return {k: v for k, v in data.items() if v and str(v).lower() != "null"}
         except Exception:
             return {}