Update app.py
Browse files
app.py
CHANGED
@@ -12,13 +12,12 @@ from langchain.prompts import PromptTemplate
|
|
12 |
from langchain_openai import ChatOpenAI
|
13 |
|
14 |
|
15 |
-
|
16 |
os.environ["OPENAI_API_KEY"] = os.environ["OPENROUTER_API_KEY"]
|
17 |
os.environ["OPENAI_API_BASE"] = "https://openrouter.ai/api/v1"
|
18 |
os.environ["OPENAI_API_HEADERS"] = '{"HTTP-Referer":"https://huggingface.co", "X-Title":"PDF-RAG"}'
|
19 |
|
20 |
-
|
21 |
-
#Section-aware PDF extractor
|
22 |
def extract_clean_sections(file_path):
|
23 |
with pdfplumber.open(file_path) as pdf:
|
24 |
full_text = ""
|
@@ -40,8 +39,7 @@ def extract_clean_sections(file_path):
|
|
40 |
docs.append(Document(page_content=f"{title}:\n{content}", metadata={"section": title}))
|
41 |
return docs
|
42 |
|
43 |
-
|
44 |
-
#TF-IDF Embedding for RAG
|
45 |
class TfidfEmbedding(Embeddings):
|
46 |
def __init__(self):
|
47 |
self.vectorizer = TfidfVectorizer()
|
@@ -55,8 +53,7 @@ class TfidfEmbedding(Embeddings):
|
|
55 |
def embed_query(self, text):
|
56 |
return self.vectorizer.transform([text]).toarray()[0]
|
57 |
|
58 |
-
|
59 |
-
# prompt
|
60 |
TEMPLATE = """
|
61 |
You are a strict healthcare policy checker for Systems Ltd.
|
62 |
|
@@ -72,17 +69,12 @@ Use the following policy information to support your answer.
|
|
72 |
Question: {question}
|
73 |
Answer:
|
74 |
"""
|
75 |
-
|
76 |
custom_prompt = PromptTemplate(template=TEMPLATE, input_variables=["context", "question"])
|
77 |
|
78 |
-
# Global state
|
79 |
-
retriever = None
|
80 |
-
qa_chain = None
|
81 |
-
|
82 |
|
83 |
-
# β
|
84 |
-
def
|
85 |
-
global
|
86 |
docs = extract_clean_sections("healthcare_policy.pdf")
|
87 |
texts = [doc.page_content for doc in docs]
|
88 |
embedder = TfidfEmbedding()
|
@@ -101,7 +93,7 @@ def load_policy():
|
|
101 |
temperature=0.0
|
102 |
)
|
103 |
|
104 |
-
|
105 |
llm=llm,
|
106 |
chain_type="stuff",
|
107 |
retriever=retriever,
|
@@ -109,27 +101,23 @@ def load_policy():
|
|
109 |
chain_type_kwargs={"prompt": custom_prompt}
|
110 |
)
|
111 |
|
112 |
-
|
113 |
-
return "Policy loaded. You may now ask questions."
|
114 |
-
|
115 |
-
|
116 |
-
# β
Answer a claim question
|
117 |
def ask_policy_question(question):
|
118 |
if qa_chain is None:
|
119 |
-
return "
|
120 |
try:
|
121 |
return qa_chain.run(question)
|
122 |
except Exception as e:
|
123 |
return f"Error: {str(e)}"
|
124 |
|
125 |
|
126 |
-
# β
Gradio
|
|
|
|
|
|
|
127 |
with gr.Blocks() as demo:
|
128 |
gr.Markdown("## SL HealthCare Claim Checker (RAG)")
|
129 |
-
|
130 |
-
load_btn = gr.Button("π₯ Ask about claim (Load Policy)")
|
131 |
-
load_status = gr.Textbox(label="Status")
|
132 |
-
load_btn.click(fn=load_policy, outputs=load_status)
|
133 |
|
134 |
with gr.Row():
|
135 |
question = gr.Textbox(label="Enter your claim question")
|
@@ -138,4 +126,13 @@ with gr.Blocks() as demo:
|
|
138 |
answer = gr.Textbox(label="Answer", lines=6)
|
139 |
ask_btn.click(fn=ask_policy_question, inputs=question, outputs=answer)
|
140 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
141 |
demo.launch()
|
|
|
12 |
from langchain_openai import ChatOpenAI
|
13 |
|
14 |
|
15 |
+
# β
OpenRouter API setup (use Hugging Face Secret)
|
16 |
os.environ["OPENAI_API_KEY"] = os.environ["OPENROUTER_API_KEY"]
|
17 |
os.environ["OPENAI_API_BASE"] = "https://openrouter.ai/api/v1"
|
18 |
os.environ["OPENAI_API_HEADERS"] = '{"HTTP-Referer":"https://huggingface.co", "X-Title":"PDF-RAG"}'
|
19 |
|
20 |
+
# β
Load and clean the policy PDF
|
|
|
21 |
def extract_clean_sections(file_path):
|
22 |
with pdfplumber.open(file_path) as pdf:
|
23 |
full_text = ""
|
|
|
39 |
docs.append(Document(page_content=f"{title}:\n{content}", metadata={"section": title}))
|
40 |
return docs
|
41 |
|
42 |
+
# β
TF-IDF Embeddings
|
|
|
43 |
class TfidfEmbedding(Embeddings):
|
44 |
def __init__(self):
|
45 |
self.vectorizer = TfidfVectorizer()
|
|
|
53 |
def embed_query(self, text):
|
54 |
return self.vectorizer.transform([text]).toarray()[0]
|
55 |
|
56 |
+
# β
Prompt Template (no emojis, no markdown)
|
|
|
57 |
TEMPLATE = """
|
58 |
You are a strict healthcare policy checker for Systems Ltd.
|
59 |
|
|
|
69 |
Question: {question}
|
70 |
Answer:
|
71 |
"""
|
|
|
72 |
custom_prompt = PromptTemplate(template=TEMPLATE, input_variables=["context", "question"])
|
73 |
|
|
|
|
|
|
|
|
|
74 |
|
75 |
+
# β
Load the policy at startup
|
76 |
+
def initialize_policy():
|
77 |
+
global qa_chain
|
78 |
docs = extract_clean_sections("healthcare_policy.pdf")
|
79 |
texts = [doc.page_content for doc in docs]
|
80 |
embedder = TfidfEmbedding()
|
|
|
93 |
temperature=0.0
|
94 |
)
|
95 |
|
96 |
+
qa_chain = RetrievalQA.from_chain_type(
|
97 |
llm=llm,
|
98 |
chain_type="stuff",
|
99 |
retriever=retriever,
|
|
|
101 |
chain_type_kwargs={"prompt": custom_prompt}
|
102 |
)
|
103 |
|
104 |
+
# β
Run QA on user question
|
|
|
|
|
|
|
|
|
105 |
def ask_policy_question(question):
|
106 |
if qa_chain is None:
|
107 |
+
return "The policy is still loading. Please wait."
|
108 |
try:
|
109 |
return qa_chain.run(question)
|
110 |
except Exception as e:
|
111 |
return f"Error: {str(e)}"
|
112 |
|
113 |
|
114 |
+
# β
Gradio Interface
|
115 |
+
qa_chain = None
|
116 |
+
status_text = "Loading..." # Initial status
|
117 |
+
|
118 |
with gr.Blocks() as demo:
|
119 |
gr.Markdown("## SL HealthCare Claim Checker (RAG)")
|
120 |
+
status_box = gr.Textbox(label="Status", value=status_text, interactive=False)
|
|
|
|
|
|
|
121 |
|
122 |
with gr.Row():
|
123 |
question = gr.Textbox(label="Enter your claim question")
|
|
|
126 |
answer = gr.Textbox(label="Answer", lines=6)
|
127 |
ask_btn.click(fn=ask_policy_question, inputs=question, outputs=answer)
|
128 |
|
129 |
+
# Load the policy on startup
|
130 |
+
def startup():
|
131 |
+
global status_text
|
132 |
+
initialize_policy()
|
133 |
+
status_text = "Policy loaded. You may now ask questions."
|
134 |
+
return status_text
|
135 |
+
|
136 |
+
demo.load(fn=startup, outputs=status_box)
|
137 |
+
|
138 |
demo.launch()
|