Update Decision Engine and DTOs for WP-25: Bump version to 1.0.3 and 0.7.1 respectively. Introduce stream tracing support by adding 'stream_origin' to QueryHit model. Enhance robustness with pre-initialization of stream variables and improve logging for unknown strategies. Refactor prompts for clarity and consistency in multi-stream synthesis.
This commit is contained in:
1 parent
5ab01c5150
commit
ea38743a2a
3 files changed
+102
-137
No files matched your search
@@ -3,12 +3,11 @@ FILE: app/core/retrieval/decision_engine.py
|
||||
DESCRIPTION: Der Agentic Orchestrator für WP-25.
|
||||
Realisiert Multi-Stream Retrieval, Intent-basiertes Routing
|
||||
und parallele Wissens-Synthese.
|
||||
VERSION: 1.0.2
|
||||
VERSION: 1.0.3
|
||||
STATUS: Active
|
||||
FIX:
|
||||
- WP-25 ROBUSTNESS: Pre-Initialization aller Stream-Variablen zur Vermeidung von KeyErrors.
|
||||
- Behebung von Template-Mismatches bei unvollständigen Strategie-Definitionen.
|
||||
- Erhalt der Sicherheitskaskade für die Strategiewahl.
|
||||
- WP-25 STREAM-TRACING: Kennzeichnung der Treffer mit ihrem Ursprungs-Stream.
|
||||
- WP-25 ROBUSTNESS: Pre-Initialization der Stream-Variablen zur Vermeidung von KeyErrors.
|
||||
"""
|
||||
import asyncio
|
||||
import logging
|
||||
@@ -50,50 +49,41 @@ class DecisionEngine:
|
||||
Hauptmethode des MindNet Chats.
|
||||
Orchestriert den gesamten Prozess: Routing -> Retrieval -> Synthese.
|
||||
"""
|
||||
# 1. Intent Recognition (Welches Werkzeug brauchen wir?)
|
||||
# 1. Intent Recognition
|
||||
strategy_key = await self._determine_strategy(query)
|
||||
|
||||
# Sicherheits-Kaskade für die Strategiewahl
|
||||
strategies = self.config.get("strategies", {})
|
||||
strategy = strategies.get(strategy_key)
|
||||
|
||||
if not strategy:
|
||||
logger.warning(f"⚠️ Unknown strategy '{strategy_key}'. Attempting fallback to FACT_WHAT.")
|
||||
logger.warning(f"⚠️ Unknown strategy '{strategy_key}'. Fallback to FACT_WHAT.")
|
||||
strategy_key = "FACT_WHAT"
|
||||
strategy = strategies.get("FACT_WHAT")
|
||||
|
||||
# WP-25 FIX: Wenn FACT_WHAT ebenfalls fehlt, wähle die erste verfügbare Strategie
|
||||
if not strategy and strategies:
|
||||
strategy_key = next(iter(strategies))
|
||||
strategy = strategies[strategy_key]
|
||||
logger.warning(f"⚠️ 'FACT_WHAT' missing in config. Using first available: {strategy_key}")
|
||||
|
||||
# Letzte Rettung: Falls gar keine Strategien definiert sind
|
||||
if not strategy:
|
||||
logger.error("❌ CRITICAL: No strategies defined in decision_engine.yaml!")
|
||||
return "Entschuldigung, meine Wissensbasis ist aktuell nicht konfiguriert."
|
||||
|
||||
# 2. Multi-Stream Retrieval (Wissen parallel sammeln)
|
||||
# 2. Multi-Stream Retrieval
|
||||
stream_results = await self._execute_parallel_streams(strategy, query)
|
||||
|
||||
# 3. Synthese (Ergebnisse zu einer Antwort verweben)
|
||||
# 3. Synthese
|
||||
return await self._generate_final_answer(strategy_key, strategy, query, stream_results)
|
||||
|
||||
async def _determine_strategy(self, query: str) -> str:
|
||||
"""Nutzt den LLM-Router zur dynamischen Wahl der Such-Strategie."""
|
||||
"""Nutzt den LLM-Router zur Wahl der Such-Strategie."""
|
||||
prompt_key = self.config.get("settings", {}).get("router_prompt_key", "intent_router_v1")
|
||||
|
||||
router_prompt_template = self.llm_service.get_prompt(prompt_key)
|
||||
if not router_prompt_template:
|
||||
return "FACT_WHAT"
|
||||
|
||||
full_prompt = router_prompt_template.format(query=query)
|
||||
|
||||
try:
|
||||
response = await self.llm_service.generate_raw_response(
|
||||
full_prompt,
|
||||
max_retries=1,
|
||||
priority="realtime"
|
||||
full_prompt, max_retries=1, priority="realtime"
|
||||
)
|
||||
return str(response).strip().upper()
|
||||
except Exception as e:
|
||||
@@ -101,13 +91,12 @@ class DecisionEngine:
|
||||
return "FACT_WHAT"
|
||||
|
||||
async def _execute_parallel_streams(self, strategy: Dict, query: str) -> Dict[str, str]:
|
||||
"""Führt alle in der Strategie definierten Such-Streams gleichzeitig aus."""
|
||||
"""Führt Such-Streams gleichzeitig aus."""
|
||||
stream_keys = strategy.get("use_streams", [])
|
||||
library = self.config.get("streams_library", {})
|
||||
|
||||
tasks = []
|
||||
active_streams = []
|
||||
|
||||
for key in stream_keys:
|
||||
stream_cfg = library.get(key)
|
||||
if stream_cfg:
|
||||
@@ -127,7 +116,10 @@ class DecisionEngine:
|
||||
return mapped_results
|
||||
|
||||
async def _run_single_stream(self, name: str, cfg: Dict, query: str) -> QueryResponse:
|
||||
"""Bereitet eine spezialisierte Suche für einen Stream vor und führt sie aus."""
|
||||
"""
|
||||
Bereitet eine spezialisierte Suche vor.
|
||||
WP-25: Taggt die Treffer mit ihrem Ursprungs-Stream.
|
||||
"""
|
||||
transformed_query = cfg.get("query_template", "{query}").format(query=query)
|
||||
|
||||
request = QueryRequest(
|
||||
@@ -139,10 +131,18 @@ class DecisionEngine:
|
||||
explain=True
|
||||
)
|
||||
|
||||
return await self.retriever.search(request)
|
||||
# Retrieval ausführen
|
||||
response = await self.retriever.search(request)
|
||||
|
||||
# WP-25: STREAM-TRACING
|
||||
# Markiere jeden Treffer mit dem Namen des Quell-Streams
|
||||
for hit in response.results:
|
||||
hit.stream_origin = name
|
||||
|
||||
return response
|
||||
|
||||
def _format_stream_context(self, response: QueryResponse) -> str:
|
||||
"""Wandelt QueryHits in einen kompakten String für das LLM um."""
|
||||
"""Wandelt QueryHits in Kontext-Strings um."""
|
||||
if not response.results:
|
||||
return "Keine spezifischen Informationen in diesem Stream gefunden."
|
||||
|
||||
@@ -161,56 +161,43 @@ class DecisionEngine:
|
||||
query: str,
|
||||
stream_results: Dict[str, str]
|
||||
) -> str:
|
||||
"""Führt die Multi-Stream Synthese durch."""
|
||||
"""Führt die Synthese durch."""
|
||||
provider = strategy.get("preferred_provider") or self.settings.MINDNET_LLM_PROVIDER
|
||||
template_key = strategy.get("prompt_template", "rag_template")
|
||||
|
||||
template = self.llm_service.get_prompt(template_key, provider=provider)
|
||||
system_prompt = self.llm_service.get_prompt("system_prompt", provider=provider)
|
||||
|
||||
# WP-25 ROBUSTNESS FIX:
|
||||
# Wir stellen sicher, dass alle Variablen, die im Template vorkommen könnten,
|
||||
# zumindest mit einem leeren String initialisiert sind.
|
||||
# WP-25 ROBUSTNESS: Pre-Initialization
|
||||
all_possible_streams = ["values_stream", "facts_stream", "biography_stream", "risk_stream", "tech_stream"]
|
||||
template_vars = {s: "" for s in all_possible_streams}
|
||||
|
||||
# Überschreiben mit tatsächlichen Ergebnissen
|
||||
template_vars.update(stream_results)
|
||||
template_vars["query"] = query
|
||||
|
||||
prepend = strategy.get("prepend_instruction", "")
|
||||
|
||||
try:
|
||||
# Sicherheitscheck: Sind alle benötigten Platzhalter im Template vorhanden?
|
||||
final_prompt = template.format(**template_vars)
|
||||
if prepend:
|
||||
final_prompt = f"{prepend}\n\n{final_prompt}"
|
||||
|
||||
response = await self.llm_service.generate_raw_response(
|
||||
final_prompt,
|
||||
system=system_prompt,
|
||||
provider=provider,
|
||||
priority="realtime"
|
||||
final_prompt, system=system_prompt, provider=provider, priority="realtime"
|
||||
)
|
||||
|
||||
if not response or len(response.strip()) < 5:
|
||||
return await self.llm_service.generate_raw_response(
|
||||
final_prompt,
|
||||
system=system_prompt,
|
||||
provider="ollama",
|
||||
priority="realtime"
|
||||
final_prompt, system=system_prompt, provider="ollama", priority="realtime"
|
||||
)
|
||||
|
||||
return response
|
||||
|
||||
except KeyError as e:
|
||||
logger.error(f"Template Variable mismatch in '{template_key}': Missing {e}")
|
||||
# Fallback: Einfaches Aneinanderreihen der gefundenen Stream-Inhalte
|
||||
fallback_context = "\n\n".join([v for v in stream_results.values() if v])
|
||||
return await self.llm_service.generate_raw_response(
|
||||
f"Beantworte: {query}\n\nKontext:\n{fallback_context}",
|
||||
system=system_prompt,
|
||||
priority="realtime"
|
||||
system=system_prompt, priority="realtime"
|
||||
)
|
||||
except Exception as e:
|
||||
logger.error(f"Final Synthesis failed: {e}")
|
||||
|
||||
+6
-2
@@ -1,7 +1,7 @@
|
||||
"""
|
||||
FILE: app/models/dto.py
|
||||
DESCRIPTION: Pydantic-Modelle (DTOs) für Request/Response Bodies. Definiert das API-Schema.
|
||||
VERSION: 0.7.0 (WP-25: Multi-Stream & Agentic RAG Support)
|
||||
VERSION: 0.7.1 (WP-25: Stream-Tracing Support)
|
||||
STATUS: Active
|
||||
DEPENDENCIES: pydantic, typing, uuid
|
||||
"""
|
||||
@@ -122,7 +122,10 @@ class Explanation(BaseModel):
|
||||
# --- Response Models ---
|
||||
|
||||
class QueryHit(BaseModel):
|
||||
"""Einzelnes Trefferobjekt."""
|
||||
"""
|
||||
Einzelnes Trefferobjekt.
|
||||
WP-25: stream_origin hinzugefügt für Tracing und Feedback-Optimierung.
|
||||
"""
|
||||
node_id: str
|
||||
note_id: str
|
||||
semantic_score: float
|
||||
@@ -133,6 +136,7 @@ class QueryHit(BaseModel):
|
||||
source: Optional[Dict] = None
|
||||
payload: Optional[Dict] = None
|
||||
explanation: Optional[Explanation] = None
|
||||
stream_origin: Optional[str] = Field(None, description="Name des Ursprungs-Streams")
|
||||
|
||||
|
||||
class QueryResponse(BaseModel):
|
||||
|
||||
Reference in new issue
Block a user