refactor: modernize data ingestion, update LLM pipeline implementation, and improve Qdrant configuration handling

app.py

@@ -1,52 +1,47 @@
-import gradio as gr
-import json
 import os
+import json
+import gradio as gr
+import requests
 
 from qdrant_client import QdrantClient
-from qdrant_client import models
-
+from qdrant_client.http import models
 from langchain_qdrant import QdrantVectorStore
 from langchain_huggingface import HuggingFaceEmbeddings
 
-from transformers import pipeline
-
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
+import torch
 
-# ---------------------------
-# Load data
-# ---------------------------
-def load_files():
-    with open("policies.json", "r", encoding="utf-8") as f:
-        policies = json.load(f)
 
-    with open("products.json", "r", encoding="utf-8") as f:
-        products = json.load(f)
+# =====================
+# CONFIG
+# =====================
+COLLECTION_NAME = "playmobil"
 
-    return policies, products
+QDRANT_URL = os.environ["QDRANT_URL"]
+QDRANT_API_KEY = os.environ["QDRANT_API_KEY"]
 
 
-# ---------------------------
-# Embeddings (local)
-# ---------------------------
+# =====================
+# EMBEDDINGS
+# =====================
 embeddings = HuggingFaceEmbeddings(
     model_name="sentence-transformers/all-MiniLM-L6-v2"
 )
 
 
-# ---------------------------
-# Qdrant client
-# ---------------------------
+# =====================
+# QDRANT CLIENT
+# =====================
 client = QdrantClient(
-    url=os.environ["QDRANT_URL"],
-    api_key=os.environ["QDRANT_API_KEY"]
+    url=QDRANT_URL,
+    api_key=QDRANT_API_KEY
 )
 
-collection_name = "playmobil"
 
-
-# create collection safely
-if not client.collection_exists(collection_name):
+# create collection safely (FIX for your 404)
+if not client.collection_exists(COLLECTION_NAME):
     client.create_collection(
-        collection_name=collection_name,
+        collection_name=COLLECTION_NAME,
         vectors_config=models.VectorParams(
             size=384,
             distance=models.Distance.COSINE
@@ -54,22 +49,52 @@ if not client.collection_exists(collection_name):
     )
 
 
-# ---------------------------
-# Vector DB
-# ---------------------------
 db = QdrantVectorStore(
     client=client,
-    collection_name=collection_name,
+    collection_name=COLLECTION_NAME,
     embedding=embeddings
 )
 
 
-# ---------------------------
-# Local LLM (NO API)
-# ---------------------------
-llm = pipeline(
+# =====================
+# LOAD DATA
+# =====================
+def load_files():
+    with open("policies.json", "r", encoding="utf-8") as f:
+        policies = json.load(f)
+
+    with open("products.json", "r", encoding="utf-8") as f:
+        products = json.load(f)
+
+    return policies, products
+
+
+def ingest_to_qdrant():
+    policies, products = load_files()
+    all_items = policies + products
+
+    texts = [item["text"] for item in all_items]
+    ids = [item["id"] for item in all_items]
+
+    db.add_texts(texts=texts, ids=ids)
+
+    return f"Ingested {len(texts)} items successfully"
+
+
+# =====================
+# LLM (FIXED - no HF API)
+# =====================
+
+model_name = "google/flan-t5-base"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+
+llm_pipe = pipeline(
     "text2text-generation",
-    model="google/flan-t5-base"
+    model=model,
+    tokenizer=tokenizer,
+    device=0 if torch.cuda.is_available() else -1
 )
 
 
@@ -83,55 +108,28 @@ Context:
 Question:
 {question}
 
-Answer:
+Answer clearly and briefly:
 """
-    result = llm(prompt, max_new_tokens=200)
-    return result[0]["generated_text"]
 
+    result = llm_pipe(prompt, max_new_tokens=200)[0]["generated_text"]
+    return result
 
-# ---------------------------
-# Ingest data
-# ---------------------------
-def ingest_to_qdrant():
-    policies, products = load_files()
-    all_items = policies + products
-
-    texts = []
-    ids = []
-    metadatas = []
-
-    for item in all_items:
-        texts.append(item["text"])
-        ids.append(item["id"])
-        metadatas.append(
-            {"type": "policy" if "policy" in item["id"] else "product"}
-        )
 
-    db.add_texts(
-        texts=texts,
-        ids=ids,
-        metadatas=metadatas
-    )
-
-    return f"Inserted {len(texts)} items into Qdrant"
-
-
-# ---------------------------
-# Chat function
-# ---------------------------
+# =====================
+# RAG CHAT
+# =====================
 def chat(question):
     docs = db.similarity_search(question, k=3)
-
     context = "\n".join([d.page_content for d in docs])
 
     return call_llm(context, question)
 
 
-# ---------------------------
-# UI (Gradio)
-# ---------------------------
+# =====================
+# UI
+# =====================
 with gr.Blocks() as demo:
-    gr.Markdown("## 🧸 Playmobil RAG Assistant (Local)")
+    gr.Markdown("# Playmobil RAG Assistant")
 
     question = gr.Textbox(label="Ask a question")
     answer = gr.Textbox(label="Answer")