feat: update embeddings, prompt, and add streaming API

src/llm/main.py

@@ -9,21 +9,25 @@ from langchain_community.embeddings import HuggingFaceEmbeddings
 
 # --- Configuration (Same as before) ---
 DB_PATH = "dune_db"
-EMBEDDING_MODEL_NAME = "all-MiniLM-L6-v2"
+EMBEDDING_MODEL_NAME = "nomic-ai/nomic-embed-text-v1.5"
 OLLAMA_API_URL = "http://localhost:11434/api/generate"
 OLLAMA_MODEL = "llama3:8b"
 PROMPT_TEMPLATE = """
-You are a helpful AI assistant and an expert on the Dune book series.
-Use the following pieces of context from the books to answer the user's question.
+You are an expert lore master for the Dune universe.
+Your task is to answer the user's question with as much detail and context as possible, based *only* on the provided text excerpts.
 If you don't know the answer from the context provided, just say that you don't know, don't try to make up an answer.
 
-Context:
+Combine all the relevant information from the context below into a single, cohesive, and comprehensive answer.
+Do not break the answer into sections based on the source texts. Synthesize them.
+The answer should be thorough and well-explained.
+
+CONTEXT:
 {context}
 
-Question:
+QUESTION:
 {question}
 
-Answer:
+ANSWER:
 """
 
 # --- Pydantic Models (Same as before) ---
@@ -34,7 +38,7 @@ class AskRequest(BaseModel):
 app = FastAPI()
 embeddings = HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL_NAME, model_kwargs={'trust_remote_code': True})
 vector_store = Chroma(persist_directory=DB_PATH, embedding_function=embeddings)
-retriever = vector_store.as_retriever(search_kwargs={"k": 5})
+retriever = vector_store.as_retriever(search_kwargs={"k": 8})
 
 # --- NEW: The Streaming Endpoint ---
 @app.post("/ask-stream")