Update app.py

app.py

@@ -1,128 +1,103 @@
+# app.py — Minimal RAG over ./data/insurance.pdf with LlamaIndex + Pinecone
+
 import os
 import logging
-import tempfile
-from typing import List
-
 import gradio as gr
 
-# ---- LlamaIndex / Pinecone ----
+# ---- Vector + LLM stack ----
 from pinecone import Pinecone, ServerlessSpec
 from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, StorageContext, Settings
 from llama_index.vector_stores.pinecone import PineconeVectorStore
 from llama_index.embeddings.openai import OpenAIEmbedding
 from llama_index.llms.openai import OpenAI
 
-# ---- Config ----
+# ========== CONFIG ==========
 PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")
-OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
+OPENAI_API_KEY   = os.getenv("OPENAI_API_KEY")
 
-# You can override these via Space "Variables" (Secrets)
-PINECONE_INDEX_NAME = os.getenv("PINECONE_INDEX_NAME", "dds-demo-index")
-PINECONE_REGION = os.getenv("PINECONE_REGION", "us-east-1")  # keep in sync with ServerlessSpec
-PINECONE_CLOUD = os.getenv("PINECONE_CLOUD", "aws")
-EMBED_MODEL = os.getenv("EMBED_MODEL", "text-embedding-3-small")
-LLM_MODEL = os.getenv("LLM_MODEL", "gpt-4o-mini")
+# Optional overrides via Space Variables
+PINECONE_INDEX_NAME = os.getenv("PINECONE_INDEX_NAME", "dds-insurance-index")
+PINECONE_REGION     = os.getenv("PINECONE_REGION", "us-east-1")
+PINECONE_CLOUD      = os.getenv("PINECONE_CLOUD", "aws")
+EMBED_MODEL         = os.getenv("EMBED_MODEL", "text-embedding-3-small")  # 1536 dims
+LLM_MODEL           = os.getenv("LLM_MODEL", "gpt-4o-mini")
+
+DATA_DIR = "data"  # place insurance.pdf inside this folder
 
 if not PINECONE_API_KEY:
-    raise RuntimeError("Missing PINECONE_API_KEY. Add it in your Space settings (Secrets).")
+    raise RuntimeError("Missing PINECONE_API_KEY (set it in your Space → Settings → Variables).")
 if not OPENAI_API_KEY:
-    raise RuntimeError("Missing OPENAI_API_KEY. Add it in your Space settings (Secrets).")
+    raise RuntimeError("Missing OPENAI_API_KEY (set it in your Space → Settings → Variables).")
 
 logging.basicConfig(level=logging.INFO)
-logger = logging.getLogger("dds-space")
+log = logging.getLogger("dds-space")
+
+# ========== CLIENTS / GLOBALS ==========
+# LlamaIndex global settings
+Settings.embed_model = OpenAIEmbedding(model=EMBED_MODEL, api_key=OPENAI_API_KEY)
+Settings.llm = OpenAI(model=LLM_MODEL, api_key=OPENAI_API_KEY)
 
-# ---- Pinecone client & index bootstrap ----
+# Pinecone
 pc = Pinecone(api_key=PINECONE_API_KEY)
 
-# Create index if it doesn't exist.
-def _ensure_index(index_name: str, dimension: int = 1536):
-    existing = [idx["name"] for idx in pc.list_indexes()]
-    if index_name not in existing:
-        logger.info(f"Creating Pinecone index '{index_name}' (dim={dimension})...")
+def ensure_index(name: str, dim: int = 1536):
+    names = [i["name"] for i in pc.list_indexes()]
+    if name not in names:
+        log.info(f"Creating Pinecone index '{name}' (dim={dim})...")
         pc.create_index(
-            name=index_name,
-            dimension=dimension,
+            name=name,
+            dimension=dim,
             metric="cosine",
             spec=ServerlessSpec(cloud=PINECONE_CLOUD, region=PINECONE_REGION),
         )
-    return pc.Index(index_name)
+    return pc.Index(name)
 
-pinecone_index = _ensure_index(PINECONE_INDEX_NAME, dimension=1536)
+pinecone_index = ensure_index(PINECONE_INDEX_NAME, dim=1536)
+vector_store = PineconeVectorStore(pinecone_index=pinecone_index)
 
-# ---- LlamaIndex settings ----
-# Set global settings for LlamaIndex (embeddings + LLM)
-Settings.embed_model = OpenAIEmbedding(model=EMBED_MODEL, api_key=OPENAI_API_KEY)
-Settings.llm = OpenAI(model=LLM_MODEL, api_key=OPENAI_API_KEY)
+# Build once on startup if index is empty (idempotent — safe to re-run)
+def bootstrap_index():
+    # If you want a quick “is empty” check, you can skip or keep this; many set-ups
+    # just upsert blindly (Pinecone dedup keys if you supply your own ids).
+    log.info("Loading documents from ./data ...")
+    if not os.path.isdir(DATA_DIR):
+        raise RuntimeError("No 'data/' directory found. Create it and add insurance.pdf.")
 
-# Vector store wrapper
-vector_store = PineconeVectorStore(pinecone_index=pinecone_index)
+    # Read everything in ./data (PDF/TXT/DOCX supported by LlamaIndex readers)
+    docs = SimpleDirectoryReader(DATA_DIR).load_data()
 
-def build_or_update_index(files: List[gr.File]) -> str:
-    """
-    Load the uploaded files, chunk them with LlamaIndex, and upsert into Pinecone.
-    """
-    if not files:
-        return "Please upload at least one file."
-    with tempfile.TemporaryDirectory() as tmpdir:
-        paths = []
-        for f in files:
-            # Gradio File object -> save to temp path
-            dst = os.path.join(tmpdir, os.path.basename(f.name))
-            with open(f.name, "rb") as src, open(dst, "wb") as out:
-                out.write(src.read())
-            paths.append(dst)
-
-        docs = SimpleDirectoryReader(input_files=paths).load_data()
-        storage_context = StorageContext.from_defaults(vector_store=vector_store)
-
-        # Build a new index (will upsert into Pinecone via the vector_store)
-        _ = VectorStoreIndex.from_documents(
-            docs,
-            storage_context=storage_context,
-            show_progress=True,
-        )
+    log.info(f"Docs loaded: {len(docs)}. Upserting into Pinecone…")
+    storage_ctx = StorageContext.from_defaults(vector_store=vector_store)
+
+    # Creates a VectorStoreIndex that writes directly to Pinecone
+    VectorStoreIndex.from_documents(docs, storage_context=storage_ctx, show_progress=True)
+    log.info("Index upsert complete.")
 
-    return f"Indexed {len(files)} file(s) into Pinecone index: {PINECONE_INDEX_NAME}."
+# Initialize the index once at app start
+bootstrap_index()
 
+# Lightweight query function (wraps the existing vector store)
 def answer(query: str, top_k: int = 4) -> str:
-    if not query or not query.strip():
-        return "Ask a question about your uploaded knowledge."
-    # Re-build a lightweight index wrapper that reads from the existing vector store
+    if not query.strip():
+        return "Please enter a question about the insurance document."
     index = VectorStoreIndex.from_vector_store(vector_store)
-    qe = index.as_query_engine(similarity_top_k=top_k)
-    resp = qe.query(query)
+    engine = index.as_query_engine(similarity_top_k=top_k)
+    resp = engine.query(query)
     return str(resp)
 
-# ---- UI ----
-INTRO = (
-    "Upload PDFs/TXT/Docs to build a Pinecone vector index (1536-d). "
-    "Then ask questions to retrieve & summarize with LlamaIndex + OpenAI."
-)
-
+# ========== UI ==========
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.Markdown("<h1 style='text-align:center;'>Insurance Q&A (RAG)</h1>")
     gr.Markdown(
-        "<h1 style='text-align:center;'>📚 RAG with LlamaIndex + Pinecone</h1>"
-        "<p style='text-align:center;'>Omantel/DDS demo Space — minimal, production-friendly layout</p>"
+        "This app indexes the file(s) in <code>./data</code> (e.g., <b>insurance.pdf</b>) "
+        "into Pinecone, then answers questions using LlamaIndex + OpenAI."
     )
+    q = gr.Textbox(label="Ask a question", placeholder="e.g., What is covered under outpatient benefits?")
+    topk = gr.Slider(1, 10, value=4, step=1, label="Top-K matches")
+    btn = gr.Button("Ask")
+    out = gr.Markdown()
 
-    with gr.Row():
-        with gr.Column(scale=1):
-            gr.Markdown("### 1) Upload & Index")
-            file_uploader = gr.File(label="Upload documents", file_count="multiple", type="filepath")
-            index_btn = gr.Button("Build / Update Index")
-            index_status = gr.Markdown()
-
-        with gr.Column(scale=1):
-            gr.Markdown("### 2) Ask a Question")
-            query = gr.Textbox(label="Your question", placeholder="e.g., What is the refund policy?")
-            topk = gr.Slider(1, 10, value=4, step=1, label="Top-K")
-            ask_btn = gr.Button("Ask")
-            answer_box = gr.Markdown()
-
-    gr.Markdown(f"**How it works:** {INTRO}")
-
-    index_btn.click(build_or_update_index, inputs=[file_uploader], outputs=[index_status])
-    ask_btn.click(answer, inputs=[query, topk], outputs=[answer_box])
+    btn.click(answer, inputs=[q, topk], outputs=[out])
 
 if __name__ == "__main__":
     demo.launch()