update vars, parameters

main.py

@@ -4,7 +4,7 @@ from dotenv import load_dotenv
 from langchain_classic.retrievers import ContextualCompressionRetriever
 from langchain_core.output_parsers import StrOutputParser
 from langchain_core.prompts import ChatPromptTemplate
-from langchain_core.runnables import RunnablePassthrough
+from langchain_core.runnables import RunnableLambda, RunnablePassthrough
 from langchain_google_community import VertexAISearchRetriever
 from langchain_google_community.vertex_rank import VertexAIRank
 from langchain_google_genai import ChatGoogleGenerativeAI
@@ -16,39 +16,62 @@ DATA_STORE = "akern-ds_1771234036654"
 MODEL = "gemini-2.5-flash"
 LOCATION = "eu"
 
+# LLM CONFIG
+TOP_K = 40
+TOP_P = 1
+TEMPERATURE = 0.0
+MAX_OUTPUT_TOKENS = 65535
+RETRIEVER_MAX_DOCS = 50
+RERANKER_MAX_RESULTS = 25
+
 with open("prompt.md") as f:
     template = f.read()
 
 prompt = ChatPromptTemplate.from_template(template)
 
 
-def format_docs(docs):
-    return "\n\n".join(doc.page_content for doc in docs)
+def format_docs(question: str) -> str:
+    retrieved_docs = base_retriever.invoke(question)
+
+    print("========== RETRIEVER DOCUMENTS ==========")
+    for idx, doc in enumerate(retrieved_docs, start=1):
+        snippet = doc.page_content[:200].replace("\n", " ")
+        print(f"[{idx}] metadata={doc.metadata['source']} | snippet=...{snippet}...")
+
+    reranked_docs = compression_retriever.invoke(question)
+    print("========== RERANKED DOCUMENTS ==========")
+    for idx, doc in enumerate(reranked_docs, start=1):
+        snippet = doc.page_content[:200].replace("\n", " ")
+        print(
+            f"[{idx}] metadata={doc.metadata['relevance_score']} | snippet=...{snippet}..."
+        )
+
+    return "\n\n".join(doc.page_content for doc in reranked_docs)
 
 
 llm = ChatGoogleGenerativeAI(
     model=MODEL,
     project=PROJECT,
     vertexai=True,
-    top_p=0.95,
-    top_k=40,
-    temperature=0.0,
-    max_output_tokens=65535,
+    top_p=TOP_P,
+    top_k=TOP_K,
+    temperature=TEMPERATURE,
+    max_output_tokens=MAX_OUTPUT_TOKENS,
 )
 
 base_retriever = VertexAISearchRetriever(
     project_id=PROJECT,
     data_store_id=DATA_STORE,
-    max_documents=50,
+    max_documents=RETRIEVER_MAX_DOCS,
     location_id=LOCATION,
     beta=True,
 )
 
 reranker = VertexAIRank(
     project_id=PROJECT,
-    location_id="global",
+    location_id="eu",
     ranking_config="default_ranking_config",
-    top_n=5,
+    top_n=RERANKER_MAX_RESULTS,
 )
 
 compression_retriever = ContextualCompressionRetriever(
@@ -56,44 +79,47 @@ compression_retriever = ContextualCompressionRetriever(
 )
 
 rag_chain = (
-    {"context": compression_retriever | format_docs, "question": RunnablePassthrough()}
+    {"context": RunnableLambda(format_docs), "question": RunnablePassthrough()}
     | prompt
     | llm
     | StrOutputParser()
 )
 
 
-def answer_questions() -> None:
-    QUESTIONS_DIR = "domande"
+# def parse_questions():
+#     domande_dir = "domande"
+#     risposte_dir = "risposte"
 
-    if not os.path.exists(QUESTIONS_DIR):
-        print(f"Errore: la directory '{QUESTIONS_DIR}' non esiste.")
-        return
+#     os.makedirs(risposte_dir, exist_ok=True)
 
-    files = sorted([f for f in os.listdir(QUESTIONS_DIR) if f.endswith(".txt")])
+#     for filename in sorted(os.listdir(domande_dir)):
+#         if not filename.lower().endswith(".txt"):
+#             continue
 
-    for filename in files:
-        filepath = os.path.join(QUESTIONS_DIR, filename)
+#         domanda_path = os.path.join(domande_dir, filename)
 
-        with open(filepath, "r", encoding="utf-8") as f:
-            question_content = f.read()
+#         with open(domanda_path, "r", encoding="utf-8") as f:
+#             contents = f.read()
+#             print(f"========== DOMANDA ({domanda_path}) ==========")
+#             print(contents)
 
-        print(f"Elaborazione: {filename}...")
+#         response = rag_chain.invoke(contents)
 
-        try:
-            response = rag_chain.invoke(question_content)
+#         print("========== RISPOSTA ==========")
+#         print(response)
+#         print("\n\n")
 
-            # Genera il nome del file di risposta (es. domanda1.txt -> risposta1.txt)
-            output_filename = filename.replace("domanda", "risposta")
+#         base_name = os.path.splitext(filename)[0]
+#         suffix = "".join(ch for ch in base_name if ch.isdigit()) or base_name
+#         risposta_path = os.path.join(risposte_dir, f"risposta{suffix}.txt")
 
-            with open(output_filename, "w", encoding="utf-8") as f:
-                f.write(response)
-
-            print(f"Risposta salvata in: {output_filename}")
-        except Exception as e:
-            print(f"Errore durante l'elaborazione di {filename}: {e}")
+#         with open(risposta_path, "w", encoding="utf-8") as f:
+#             f.write(response)
 
 
 if __name__ == "__main__":
-    response = rag_chain.invoke("come si calcola il rapporto sodio potassio?")
+    response = rag_chain.invoke(
+        """Buongiorno, non so se è la mail specifica ma volevo se possibile dei chiarimenti per l’interpretazione dei parametri BCM /SMM/ASMM. Mi capita a volte di trovare casi in cui la BCM è aumentata ma allo stesso tempo SMM/ASMM hanno subito una piccola flessione in negativo (o viceversa). Se la parte metabolicamente attiva aumenta perchè può succedere che gli altri compartimenti si riducono?? E allo stesso tempo phA e BCM possono essere inversamente proporzionali?? So che il phA correla con massa e struttura + idratazione."""
+    )
+
     print(response)