Update app.py
Browse files
app.py
CHANGED
|
@@ -1,32 +1,30 @@
|
|
| 1 |
-
# app.py — Insurance Q&A (RAG) with Omantel
|
| 2 |
-
#
|
| 3 |
|
| 4 |
import os
|
| 5 |
import logging
|
| 6 |
import gradio as gr
|
| 7 |
|
| 8 |
-
# ---- Vector + LLM stack ----
|
| 9 |
from pinecone import Pinecone, ServerlessSpec
|
| 10 |
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, StorageContext, Settings
|
| 11 |
from llama_index.vector_stores.pinecone import PineconeVectorStore
|
| 12 |
from llama_index.embeddings.openai import OpenAIEmbedding
|
| 13 |
from llama_index.llms.openai import OpenAI
|
| 14 |
|
| 15 |
-
#
|
| 16 |
PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")
|
| 17 |
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
|
| 18 |
|
| 19 |
-
# Optional overrides via Space Variables
|
| 20 |
PINECONE_INDEX_NAME = os.getenv("PINECONE_INDEX_NAME", "dds-insurance-index")
|
| 21 |
PINECONE_REGION = os.getenv("PINECONE_REGION", "us-east-1")
|
| 22 |
PINECONE_CLOUD = os.getenv("PINECONE_CLOUD", "aws")
|
| 23 |
EMBED_MODEL = os.getenv("EMBED_MODEL", "text-embedding-3-small") # 1536-dim
|
| 24 |
LLM_MODEL = os.getenv("LLM_MODEL", "gpt-4o-mini")
|
| 25 |
|
| 26 |
-
DATA_DIR = "data"
|
| 27 |
-
DEFAULT_TOP_K = 4
|
| 28 |
|
| 29 |
-
# Omantel
|
| 30 |
LOGO_URL = "https://raw.githubusercontent.com/Decoding-Data-Science/Omantel/main/Omantel_Logo%20(1).png"
|
| 31 |
|
| 32 |
if not PINECONE_API_KEY:
|
|
@@ -37,12 +35,10 @@ if not OPENAI_API_KEY:
|
|
| 37 |
logging.basicConfig(level=logging.INFO)
|
| 38 |
log = logging.getLogger("dds-space")
|
| 39 |
|
| 40 |
-
#
|
| 41 |
-
# LlamaIndex global settings
|
| 42 |
Settings.embed_model = OpenAIEmbedding(model=EMBED_MODEL, api_key=OPENAI_API_KEY)
|
| 43 |
Settings.llm = OpenAI(model=LLM_MODEL, api_key=OPENAI_API_KEY)
|
| 44 |
|
| 45 |
-
# Pinecone
|
| 46 |
pc = Pinecone(api_key=PINECONE_API_KEY)
|
| 47 |
|
| 48 |
def ensure_index(name: str, dim: int = 1536):
|
|
@@ -61,25 +57,20 @@ pinecone_index = ensure_index(PINECONE_INDEX_NAME, dim=1536)
|
|
| 61 |
vector_store = PineconeVectorStore(pinecone_index=pinecone_index)
|
| 62 |
|
| 63 |
def bootstrap_index():
|
| 64 |
-
"""Index all files in ./data into Pinecone (idempotent safe)."""
|
| 65 |
if not os.path.isdir(DATA_DIR):
|
| 66 |
raise RuntimeError("No 'data/' directory found. Commit your documents to data/ in the Space repo.")
|
| 67 |
-
|
| 68 |
log.info("Loading documents from ./data ...")
|
| 69 |
docs = SimpleDirectoryReader(DATA_DIR).load_data()
|
| 70 |
if not docs:
|
| 71 |
raise RuntimeError("No documents found in data/. Add e.g., data/insurance.pdf")
|
| 72 |
-
|
| 73 |
log.info(f"Docs loaded: {len(docs)}. Upserting into Pinecone…")
|
| 74 |
storage_ctx = StorageContext.from_defaults(vector_store=vector_store)
|
| 75 |
VectorStoreIndex.from_documents(docs, storage_context=storage_ctx, show_progress=True)
|
| 76 |
log.info("Index upsert complete.")
|
| 77 |
|
| 78 |
-
# Build once at startup
|
| 79 |
bootstrap_index()
|
| 80 |
|
| 81 |
def answer(query: str) -> str:
|
| 82 |
-
"""Query the existing vector store and return an answer string."""
|
| 83 |
if not query or not query.strip():
|
| 84 |
return "Please enter a question (or select one from the FAQ list)."
|
| 85 |
index = VectorStoreIndex.from_vector_store(vector_store)
|
|
@@ -87,7 +78,6 @@ def answer(query: str) -> str:
|
|
| 87 |
resp = engine.query(query)
|
| 88 |
return str(resp)
|
| 89 |
|
| 90 |
-
# ---- Frequently Asked Questions (edit to your document) ----
|
| 91 |
FAQS = [
|
| 92 |
"",
|
| 93 |
"What benefits are covered under the policy?",
|
|
@@ -101,52 +91,30 @@ FAQS = [
|
|
| 101 |
]
|
| 102 |
|
| 103 |
def use_faq(selected_faq: str, free_text: str):
|
| 104 |
-
"""
|
| 105 |
-
If a FAQ is selected, prefer it; otherwise use free_text.
|
| 106 |
-
Returns the chosen prompt (echo in UI) and the model answer.
|
| 107 |
-
"""
|
| 108 |
prompt = (selected_faq or "").strip() or (free_text or "").strip()
|
| 109 |
if not prompt:
|
| 110 |
return "", "Please select a FAQ or type your question."
|
| 111 |
return prompt, answer(prompt)
|
| 112 |
|
| 113 |
-
#
|
| 114 |
CSS = """
|
| 115 |
-
.header {
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
justify-content: center;
|
| 120 |
-
margin-top: 8px;
|
| 121 |
-
}
|
| 122 |
-
.header img {
|
| 123 |
-
height: 42px;
|
| 124 |
-
}
|
| 125 |
-
.header h1 {
|
| 126 |
-
margin: 0;
|
| 127 |
-
font-weight: 700;
|
| 128 |
-
font-size: 1.4rem;
|
| 129 |
-
}
|
| 130 |
-
.subnote {
|
| 131 |
-
text-align: center;
|
| 132 |
-
margin-top: -6px;
|
| 133 |
-
opacity: 0.8;
|
| 134 |
-
}
|
| 135 |
"""
|
| 136 |
|
| 137 |
with gr.Blocks(css=CSS, theme=gr.themes.Soft()) as demo:
|
| 138 |
-
#
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
f"""
|
| 142 |
<div class="header">
|
| 143 |
<img src="{LOGO_URL}" alt="Omantel logo" />
|
| 144 |
-
<h1>Omantel Insurance Q&A — RAG Assistant</h1>
|
| 145 |
</div>
|
|
|
|
| 146 |
<p class="subnote">Ask about coverage, claims, exclusions, and more — powered by LlamaIndex + Pinecone</p>
|
| 147 |
-
"""
|
| 148 |
-
|
| 149 |
-
)
|
| 150 |
|
| 151 |
with gr.Row():
|
| 152 |
with gr.Column(scale=1):
|
|
@@ -159,7 +127,6 @@ with gr.Blocks(css=CSS, theme=gr.themes.Soft()) as demo:
|
|
| 159 |
placeholder="e.g., What is covered under outpatient benefits?",
|
| 160 |
lines=2
|
| 161 |
)
|
| 162 |
-
|
| 163 |
ask_btn = gr.Button("Ask", variant="primary")
|
| 164 |
|
| 165 |
with gr.Column(scale=1):
|
|
|
|
| 1 |
+
# app.py — Insurance Q&A (RAG) with Omantel logo from GitHub URL (centered top)
|
| 2 |
+
# Minimal changes; logic preserved. Uses Pinecone + LlamaIndex + OpenAI.
|
| 3 |
|
| 4 |
import os
|
| 5 |
import logging
|
| 6 |
import gradio as gr
|
| 7 |
|
|
|
|
| 8 |
from pinecone import Pinecone, ServerlessSpec
|
| 9 |
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, StorageContext, Settings
|
| 10 |
from llama_index.vector_stores.pinecone import PineconeVectorStore
|
| 11 |
from llama_index.embeddings.openai import OpenAIEmbedding
|
| 12 |
from llama_index.llms.openai import OpenAI
|
| 13 |
|
| 14 |
+
# ===== CONFIG =====
|
| 15 |
PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")
|
| 16 |
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
|
| 17 |
|
|
|
|
| 18 |
PINECONE_INDEX_NAME = os.getenv("PINECONE_INDEX_NAME", "dds-insurance-index")
|
| 19 |
PINECONE_REGION = os.getenv("PINECONE_REGION", "us-east-1")
|
| 20 |
PINECONE_CLOUD = os.getenv("PINECONE_CLOUD", "aws")
|
| 21 |
EMBED_MODEL = os.getenv("EMBED_MODEL", "text-embedding-3-small") # 1536-dim
|
| 22 |
LLM_MODEL = os.getenv("LLM_MODEL", "gpt-4o-mini")
|
| 23 |
|
| 24 |
+
DATA_DIR = "data"
|
| 25 |
+
DEFAULT_TOP_K = 4 # internal similarity_top_k (no UI control)
|
| 26 |
|
| 27 |
+
# Omantel logo (raw GitHub URL so it renders directly)
|
| 28 |
LOGO_URL = "https://raw.githubusercontent.com/Decoding-Data-Science/Omantel/main/Omantel_Logo%20(1).png"
|
| 29 |
|
| 30 |
if not PINECONE_API_KEY:
|
|
|
|
| 35 |
logging.basicConfig(level=logging.INFO)
|
| 36 |
log = logging.getLogger("dds-space")
|
| 37 |
|
| 38 |
+
# ===== LlamaIndex / Pinecone =====
|
|
|
|
| 39 |
Settings.embed_model = OpenAIEmbedding(model=EMBED_MODEL, api_key=OPENAI_API_KEY)
|
| 40 |
Settings.llm = OpenAI(model=LLM_MODEL, api_key=OPENAI_API_KEY)
|
| 41 |
|
|
|
|
| 42 |
pc = Pinecone(api_key=PINECONE_API_KEY)
|
| 43 |
|
| 44 |
def ensure_index(name: str, dim: int = 1536):
|
|
|
|
| 57 |
vector_store = PineconeVectorStore(pinecone_index=pinecone_index)
|
| 58 |
|
| 59 |
def bootstrap_index():
|
|
|
|
| 60 |
if not os.path.isdir(DATA_DIR):
|
| 61 |
raise RuntimeError("No 'data/' directory found. Commit your documents to data/ in the Space repo.")
|
|
|
|
| 62 |
log.info("Loading documents from ./data ...")
|
| 63 |
docs = SimpleDirectoryReader(DATA_DIR).load_data()
|
| 64 |
if not docs:
|
| 65 |
raise RuntimeError("No documents found in data/. Add e.g., data/insurance.pdf")
|
|
|
|
| 66 |
log.info(f"Docs loaded: {len(docs)}. Upserting into Pinecone…")
|
| 67 |
storage_ctx = StorageContext.from_defaults(vector_store=vector_store)
|
| 68 |
VectorStoreIndex.from_documents(docs, storage_context=storage_ctx, show_progress=True)
|
| 69 |
log.info("Index upsert complete.")
|
| 70 |
|
|
|
|
| 71 |
bootstrap_index()
|
| 72 |
|
| 73 |
def answer(query: str) -> str:
|
|
|
|
| 74 |
if not query or not query.strip():
|
| 75 |
return "Please enter a question (or select one from the FAQ list)."
|
| 76 |
index = VectorStoreIndex.from_vector_store(vector_store)
|
|
|
|
| 78 |
resp = engine.query(query)
|
| 79 |
return str(resp)
|
| 80 |
|
|
|
|
| 81 |
FAQS = [
|
| 82 |
"",
|
| 83 |
"What benefits are covered under the policy?",
|
|
|
|
| 91 |
]
|
| 92 |
|
| 93 |
def use_faq(selected_faq: str, free_text: str):
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
prompt = (selected_faq or "").strip() or (free_text or "").strip()
|
| 95 |
if not prompt:
|
| 96 |
return "", "Please select a FAQ or type your question."
|
| 97 |
return prompt, answer(prompt)
|
| 98 |
|
| 99 |
+
# ===== UI =====
|
| 100 |
CSS = """
|
| 101 |
+
.header { text-align:center; }
|
| 102 |
+
.header img { max-height:80px; height:auto; }
|
| 103 |
+
.title { text-align:center; font-weight:700; font-size:1.4rem; margin:6px 0 0 0; }
|
| 104 |
+
.subnote { text-align:center; margin-top:-2px; opacity:0.8; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
"""
|
| 106 |
|
| 107 |
with gr.Blocks(css=CSS, theme=gr.themes.Soft()) as demo:
|
| 108 |
+
# Centered logo + title
|
| 109 |
+
gr.Markdown(
|
| 110 |
+
f"""
|
|
|
|
| 111 |
<div class="header">
|
| 112 |
<img src="{LOGO_URL}" alt="Omantel logo" />
|
|
|
|
| 113 |
</div>
|
| 114 |
+
<h1 class="title">Omantel Insurance Q&A — RAG Assistant</h1>
|
| 115 |
<p class="subnote">Ask about coverage, claims, exclusions, and more — powered by LlamaIndex + Pinecone</p>
|
| 116 |
+
"""
|
| 117 |
+
)
|
|
|
|
| 118 |
|
| 119 |
with gr.Row():
|
| 120 |
with gr.Column(scale=1):
|
|
|
|
| 127 |
placeholder="e.g., What is covered under outpatient benefits?",
|
| 128 |
lines=2
|
| 129 |
)
|
|
|
|
| 130 |
ask_btn = gr.Button("Ask", variant="primary")
|
| 131 |
|
| 132 |
with gr.Column(scale=1):
|