feat: 🔗 Tidy

src/config.yaml

@@ -5,8 +5,9 @@ api:
 
 # --- Model Settings ---
 models:
-  inference: "lm_studio/qwen/qwen3-8b"
+  enrich: "lm_studio/qwen/qwen3-8b"
   embedding: "text-embedding-qwen3-embedding-8b"
+  retrieval: "lm_studio/qwen/qwen3-next-80b"
 
 # --- Ingestion Settings ---
 ingestion:
@@ -15,6 +16,7 @@ ingestion:
   max_workers: 8
   chunk_size: 800
   chunk_overlap: 100
+  embedding_batch_size: 32
 
 # --- Retrieval Settings ---
 retrieval:

src/experts/dnd_agent.py

@@ -3,6 +3,16 @@ from langchain_community.vectorstores import FAISS
 from embedding import LocalLMEmbeddings
 from pathlib import Path
 
+from config_loader import load_config
+
+
+CFG = load_config()
+
+DATABASE_PATH = CFG["ingestion"]["db_path"]
+EMBEDDING_MODEL = CFG["models"]["embedding"]
+API_BASE = CFG["api"]["base_url"]
+
+
 # --- DSPy Signature ---
 class DnDContextQA(dspy.Signature):
     """Answer DnD campaign questions using provided snippets and full file context.
@@ -13,12 +23,12 @@ class DnDContextQA(dspy.Signature):
 
 # --- DSPy Module ---
 class DnDRAG(dspy.Module):
-    def __init__(self, db_path="./local_faiss_db", k=3):
+    def __init__(self, db_path=DATABASE_PATH, k=3):
         super().__init__()
         # 1. Setup Embeddings & Load FAISS
         self.embeddings = LocalLMEmbeddings(
-            model="text-embedding-qwen3-embedding-8b", 
+            model=EMBEDDING_MODEL, 
-            base_url="http://192.168.0.49:1234"
+            base_url=API_BASE
         )
         self.vectorstore = FAISS.load_local(
             db_path, self.embeddings, allow_dangerous_deserialization=True

src/ingest.py

@@ -15,6 +15,15 @@ from config_loader import load_config
 
 CFG = load_config()
 DATA_DIR = CFG["ingestion"]["data_dir"]
+DATABASE_PATH = CFG["ingestion"]["db_path"]
+MODEL_BASE = CFG["models"]["enrich"]
+EMBEDDING_MODEL = CFG["models"]["embedding"]
+API_BASE = CFG["api"]["base_url"]
+API_VERSION = CFG["api"]["api_version"]
+MAX_WORKERS=CFG["ingestion"]["max_workers"]
+CHUNK_SIZE=CFG["ingestion"]["chunk_size"], 
+CHUNK_OVERLAP=CFG["ingestion"]["chunk_overlap"]
+EMBEDDING_BATCH_SIZE=CFG["ingestion"]["embedding_batch_size"]
 
 def load_documents():
     docs = []
@@ -44,16 +53,13 @@ def load_documents():
 def chunk_documents(docs):
     # LangChain preserves metadata during splitting automatically
     text_splitter = RecursiveCharacterTextSplitter(
-        chunk_size=CFG["ingestion"]["chunk_size"], 
+        chunk_size=CHUNK_SIZE,
-        chunk_overlap=CFG["ingestion"]["chunk_overlap"], 
+        chunk_overlap=CHUNK_OVERLAP,
         separators=["\n\n", "\n", ". ", " ", ""]
     )
     return text_splitter.split_documents(docs)
 
 def enrich_chunks(chunks: list) -> list:
-    MODEL_BASE = CFG["models"]["inference"]
-    API_BASE = CFG["api"]["base_url"]
-    API_VERSION = CFG["api"]["api_version"]
     
     def process_single_chunk(indexed_chunk):
         idx, chunk = indexed_chunk
@@ -75,7 +81,7 @@ def enrich_chunks(chunks: list) -> list:
             
 
     enriched_results = []
-    with ThreadPoolExecutor(max_workers=CFG["ingestion"]["max_workers"]) as executor:
+    with ThreadPoolExecutor(max_workers=MAX_WORKERS) as executor:
         # Wrap chunks in enumerate to keep track of order
         futures = [executor.submit(process_single_chunk, (i, c)) for i, c in enumerate(chunks)]
 
@@ -86,11 +92,11 @@ def enrich_chunks(chunks: list) -> list:
     enriched_results.sort(key=lambda x: x[0])
     return [item[1] for item in enriched_results]
 
-def store_chunks_locally(chunks, db_path="./local_faiss_db"):
+def store_chunks_locally(chunks, db_path=DATABASE_PATH):
     embeddings_model = LocalLMEmbeddings(
-        model="text-embedding-qwen3-embedding-8b",
+        model=EMBEDDING_MODEL,
-        base_url="http://192.168.0.49:1234",
+        base_url=API_BASE,
-        batch_size=32,
+        batch_size=EMBEDDING_BATCH_SIZE,
     )
 
     print(f"Index creation started for {len(chunks)} chunks...")

src/retrieve.py

@@ -1,11 +1,19 @@
 import sys
 import dspy
 from experts.dnd_agent import DnDRAG
+from config_loader import load_config
+
+
+CFG = load_config()
+RETRIEVE_MODEL = CFG["models"]["retrieval"]
+API_BASE = CFG["api"]["base_url"]
+API_VERSION = CFG["api"]["api_version"]
+
 
 def main():
     # 1. Setup the LLM
     print("🚀 Initializing Qwen-8B via LM Studio...")
-    lm = dspy.LM("lm_studio/qwen/qwen3-8b", api_base="http://192.168.0.49:1234/v1/")
+    lm = dspy.LM(RETRIEVE_MODEL, api_base=API_BASE+API_VERSION)
     dspy.configure(lm=lm)
 
     # 2. Load the RAG System (only happens once!)