Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -9,115 +9,191 @@ from langchain_text_splitters import RecursiveCharacterTextSplitter
|
|
| 9 |
from langchain_community.embeddings import HuggingFaceEmbeddings
|
| 10 |
from langchain_community.vectorstores import Chroma
|
| 11 |
import torch
|
| 12 |
-
import warnings
|
| 13 |
-
|
| 14 |
-
# Suppress warnings that clutter the Streamlit interface
|
| 15 |
-
warnings.filterwarnings("ignore")
|
| 16 |
-
logging.basicConfig(level=logging.INFO)
|
| 17 |
|
| 18 |
# ---------------- CONFIGURATION ----------------
|
|
|
|
| 19 |
|
| 20 |
-
# Load API key from Hugging Face secrets
|
| 21 |
GROQ_API_KEY = st.secrets.get("GROQ_API_KEY", os.environ.get("GROQ_API_KEY"))
|
| 22 |
GROQ_MODEL = "llama-3.1-8b-instant"
|
| 23 |
|
| 24 |
-
# Initialize Groq client
|
| 25 |
client = None
|
| 26 |
if GROQ_API_KEY:
|
| 27 |
try:
|
| 28 |
client = Groq(api_key=GROQ_API_KEY)
|
| 29 |
-
|
| 30 |
-
logging.info("β
Groq client initialized successfully.")
|
| 31 |
except Exception as e:
|
| 32 |
-
|
| 33 |
client = None
|
| 34 |
else:
|
|
|
|
| 35 |
st.warning("β οΈ GROQ_API_KEY not found. Please add it to Hugging Face secrets.")
|
| 36 |
|
| 37 |
# ---------------- STREAMLIT UI SETUP ----------------
|
| 38 |
st.set_page_config(page_title="PDF Assistant", page_icon="π", layout="wide")
|
| 39 |
|
| 40 |
-
# ---------------- CSS (
|
| 41 |
-
st.markdown(
|
|
|
|
| 42 |
<style>
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
overflow-y: auto;
|
|
|
|
|
|
|
| 49 |
}
|
| 50 |
|
| 51 |
-
/*
|
| 52 |
-
.
|
| 53 |
-
|
| 54 |
-
|
| 55 |
}
|
| 56 |
|
| 57 |
-
/*
|
| 58 |
-
|
| 59 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
}
|
| 61 |
-
/* ======================================================= */
|
| 62 |
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
}
|
| 69 |
|
|
|
|
| 70 |
.chat-user {
|
| 71 |
background: #2d3748;
|
| 72 |
-
padding: 12px;
|
| 73 |
-
border-radius:
|
| 74 |
-
margin:
|
| 75 |
-
max-width:
|
| 76 |
-
|
| 77 |
-
|
|
|
|
| 78 |
}
|
| 79 |
.chat-bot {
|
| 80 |
-
background: var(--primary
|
| 81 |
-
padding: 12px;
|
| 82 |
-
border-radius:
|
| 83 |
-
margin:
|
| 84 |
-
max-width:
|
| 85 |
-
|
| 86 |
-
|
|
|
|
| 87 |
}
|
| 88 |
-
|
| 89 |
.sources {
|
| 90 |
-
font-size: 0.
|
| 91 |
-
opacity: 0.
|
| 92 |
-
margin-top:
|
| 93 |
-
border-top: 1px solid rgba(255,
|
| 94 |
-
padding-top:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
}
|
| 96 |
|
| 97 |
-
/*
|
| 98 |
-
|
| 99 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
}
|
| 101 |
</style>
|
| 102 |
-
""",
|
|
|
|
|
|
|
| 103 |
|
| 104 |
# ---------------- SESSION STATE ----------------
|
| 105 |
if "chat" not in st.session_state:
|
| 106 |
st.session_state.chat = []
|
| 107 |
-
|
| 108 |
if "vectorstore" not in st.session_state:
|
| 109 |
st.session_state.vectorstore = None
|
| 110 |
-
|
| 111 |
if "retriever" not in st.session_state:
|
| 112 |
st.session_state.retriever = None
|
| 113 |
-
|
| 114 |
if "uploaded_file_name" not in st.session_state:
|
| 115 |
st.session_state.uploaded_file_name = None
|
| 116 |
-
|
| 117 |
if "uploader_key" not in st.session_state:
|
| 118 |
st.session_state.uploader_key = 0
|
| 119 |
|
| 120 |
-
# ----------------
|
| 121 |
def clear_chat_history():
|
| 122 |
st.session_state.chat = []
|
| 123 |
|
|
@@ -132,57 +208,47 @@ def clear_memory():
|
|
| 132 |
st.success("Memory cleared. Please upload a new PDF.")
|
| 133 |
|
| 134 |
def process_pdf(uploaded_file):
|
| 135 |
-
"""Process uploaded PDF and create vectorstore."""
|
| 136 |
try:
|
| 137 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
|
| 138 |
tmp.write(uploaded_file.getvalue())
|
| 139 |
path = tmp.name
|
| 140 |
-
|
| 141 |
loader = PyPDFLoader(path)
|
| 142 |
docs = loader.load()
|
| 143 |
-
|
| 144 |
-
splitter = RecursiveCharacterTextSplitter(
|
| 145 |
-
chunk_size=800,
|
| 146 |
-
chunk_overlap=50
|
| 147 |
-
)
|
| 148 |
chunks = splitter.split_documents(docs)
|
| 149 |
-
|
| 150 |
embeddings = HuggingFaceEmbeddings(
|
| 151 |
model_name="sentence-transformers/all-MiniLM-L6-v2",
|
| 152 |
model_kwargs={"device": "cpu"},
|
| 153 |
-
encode_kwargs={"normalize_embeddings": True}
|
| 154 |
)
|
| 155 |
-
|
| 156 |
vectorstore = Chroma.from_documents(chunks, embeddings)
|
| 157 |
retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
|
| 158 |
-
|
| 159 |
st.session_state.vectorstore = vectorstore
|
| 160 |
st.session_state.retriever = retriever
|
| 161 |
-
|
| 162 |
if os.path.exists(path):
|
| 163 |
os.unlink(path)
|
| 164 |
-
|
| 165 |
return len(chunks)
|
| 166 |
-
|
| 167 |
except Exception as e:
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
if 'path' in locals() and os.path.exists(path):
|
| 171 |
-
os.unlink(path)
|
| 172 |
return None
|
| 173 |
|
| 174 |
def ask_question(question):
|
| 175 |
-
"""Retrieve and generate answer for the question."""
|
| 176 |
if not client:
|
| 177 |
return None, 0, "Groq client is not initialized. Check API key setup."
|
| 178 |
-
|
| 179 |
if not st.session_state.retriever:
|
| 180 |
return None, 0, "Upload PDF first to initialize the knowledge base."
|
| 181 |
-
|
| 182 |
try:
|
| 183 |
docs = st.session_state.retriever.invoke(question)
|
| 184 |
context = "\n\n".join(d.page_content for d in docs)
|
| 185 |
-
|
| 186 |
prompt = f"""
|
| 187 |
You are a strict RAG Q&A assistant.
|
| 188 |
Use ONLY the context provided. If the answer is not found, reply:
|
|
@@ -199,120 +265,108 @@ FINAL ANSWER:
|
|
| 199 |
response = client.chat.completions.create(
|
| 200 |
model=GROQ_MODEL,
|
| 201 |
messages=[
|
| 202 |
-
{"role": "system",
|
| 203 |
-
|
| 204 |
-
{"role": "user", "content": prompt}
|
| 205 |
],
|
| 206 |
-
temperature=0.0
|
| 207 |
)
|
| 208 |
-
|
| 209 |
answer = response.choices[0].message.content.strip()
|
| 210 |
return answer, len(docs), None
|
| 211 |
-
|
| 212 |
except APIError as e:
|
| 213 |
return None, 0, f"Groq API Error: {str(e)}"
|
| 214 |
except Exception as e:
|
|
|
|
| 215 |
return None, 0, f"General error: {str(e)}"
|
| 216 |
|
| 217 |
-
|
| 218 |
-
"""Logic for processing the question, triggered by Enter or button."""
|
| 219 |
-
question = st.session_state.question_input
|
| 220 |
-
|
| 221 |
-
# Check if the question is non-empty and the input is enabled
|
| 222 |
-
if question and not (st.session_state.uploaded_file_name is None or client is None):
|
| 223 |
-
# Add user query to chat history
|
| 224 |
-
st.session_state.chat.append(("user", question))
|
| 225 |
-
|
| 226 |
-
# Get answer
|
| 227 |
-
with st.spinner("Thinking..."):
|
| 228 |
-
answer, sources, error = ask_question(question)
|
| 229 |
-
|
| 230 |
-
if answer:
|
| 231 |
-
bot_message = f"{answer}<div class='sources'>Context Chunks Used: {sources}</div>"
|
| 232 |
-
st.session_state.chat.append(("bot", bot_message))
|
| 233 |
-
else:
|
| 234 |
-
st.session_state.chat.append(("bot", f"π΄ **Error:** {error}"))
|
| 235 |
-
|
| 236 |
-
# Clear the input box after submission
|
| 237 |
-
st.session_state.question_input = ""
|
| 238 |
-
st.rerun()
|
| 239 |
-
|
| 240 |
-
# ---------------- UI COMPONENTS ----------------
|
| 241 |
-
|
| 242 |
-
# 2. Title and Creator Info (Top Left)
|
| 243 |
-
col1, col2 = st.columns([0.4, 0.6])
|
| 244 |
-
with col1:
|
| 245 |
-
st.title("π PDF Assistant")
|
| 246 |
-
st.markdown(
|
| 247 |
-
'**Creator:** [Abhishek Saxena](https://www.linkedin.com/in/abhishek-iitr/ "Connect on LinkedIn")',
|
| 248 |
-
unsafe_allow_html=True
|
| 249 |
-
)
|
| 250 |
-
with col2:
|
| 251 |
-
pass
|
| 252 |
-
|
| 253 |
-
# Sidebar Controls
|
| 254 |
with st.sidebar:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 255 |
st.header("Controls")
|
| 256 |
st.button("ποΈ Clear Chat History", on_click=clear_chat_history, use_container_width=True)
|
| 257 |
st.button("π₯ Clear PDF Memory", on_click=clear_memory, use_container_width=True)
|
| 258 |
-
|
| 259 |
st.markdown("---")
|
|
|
|
| 260 |
if st.session_state.uploaded_file_name:
|
| 261 |
-
st.success(f"β
|
| 262 |
else:
|
| 263 |
-
st.
|
| 264 |
-
|
| 265 |
-
#
|
| 266 |
-
|
| 267 |
-
|
| 268 |
-
|
| 269 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 270 |
)
|
| 271 |
|
| 272 |
-
|
| 273 |
-
|
| 274 |
-
|
| 275 |
-
|
| 276 |
-
|
| 277 |
-
|
| 278 |
-
|
| 279 |
-
if
|
| 280 |
-
st.
|
| 281 |
-
st.session_state.uploaded_file_name = uploaded.name
|
| 282 |
else:
|
| 283 |
-
st.
|
| 284 |
-
|
| 285 |
-
|
| 286 |
-
|
| 287 |
-
|
| 288 |
-
|
| 289 |
-
st.markdown("
|
| 290 |
-
|
| 291 |
-
|
| 292 |
-
|
| 293 |
-
|
| 294 |
-
|
| 295 |
-
|
| 296 |
-
|
| 297 |
-
|
| 298 |
-
|
| 299 |
-
|
| 300 |
-
|
| 301 |
-
|
| 302 |
-
|
| 303 |
-
|
| 304 |
-
|
| 305 |
-
|
| 306 |
-
|
| 307 |
-
|
| 308 |
-
|
| 309 |
-
|
| 310 |
-
|
| 311 |
-
|
| 312 |
-
|
| 313 |
-
|
| 314 |
-
|
| 315 |
-
|
| 316 |
-
|
| 317 |
-
|
| 318 |
-
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
from langchain_community.embeddings import HuggingFaceEmbeddings
|
| 10 |
from langchain_community.vectorstores import Chroma
|
| 11 |
import torch
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
# ---------------- CONFIGURATION ----------------
|
| 14 |
+
logging.basicConfig(level=logging.INFO)
|
| 15 |
|
| 16 |
+
# Load API key from Hugging Face secrets (or env)
|
| 17 |
GROQ_API_KEY = st.secrets.get("GROQ_API_KEY", os.environ.get("GROQ_API_KEY"))
|
| 18 |
GROQ_MODEL = "llama-3.1-8b-instant"
|
| 19 |
|
| 20 |
+
# Initialize Groq client silently (no top green message)
|
| 21 |
client = None
|
| 22 |
if GROQ_API_KEY:
|
| 23 |
try:
|
| 24 |
client = Groq(api_key=GROQ_API_KEY)
|
| 25 |
+
logging.info("Groq client initialized (silent).")
|
|
|
|
| 26 |
except Exception as e:
|
| 27 |
+
logging.exception("Groq init failed.")
|
| 28 |
client = None
|
| 29 |
else:
|
| 30 |
+
# Keep a visible but non-green hint if key is missing
|
| 31 |
st.warning("β οΈ GROQ_API_KEY not found. Please add it to Hugging Face secrets.")
|
| 32 |
|
| 33 |
# ---------------- STREAMLIT UI SETUP ----------------
|
| 34 |
st.set_page_config(page_title="PDF Assistant", page_icon="π", layout="wide")
|
| 35 |
|
| 36 |
+
# ---------------- CSS (layout + styling) ----------------
|
| 37 |
+
st.markdown(
|
| 38 |
+
"""
|
| 39 |
<style>
|
| 40 |
+
:root{
|
| 41 |
+
--primary:#1e3a8a;
|
| 42 |
+
--bg:#0e1117;
|
| 43 |
+
--bg2:#1a1d29;
|
| 44 |
+
--text:#f0f2f6;
|
| 45 |
+
}
|
| 46 |
+
|
| 47 |
+
/* Fixed sidebar */
|
| 48 |
+
section[data-testid="stSidebar"] {
|
| 49 |
+
position: fixed;
|
| 50 |
+
height: 100vh;
|
| 51 |
overflow-y: auto;
|
| 52 |
+
padding-top: 0.5rem;
|
| 53 |
+
width: 300px;
|
| 54 |
}
|
| 55 |
|
| 56 |
+
/* Main content offset to the right of sidebar */
|
| 57 |
+
.main {
|
| 58 |
+
margin-left: 320px;
|
| 59 |
+
padding-top: 16px;
|
| 60 |
}
|
| 61 |
|
| 62 |
+
/* Header (title + creator) */
|
| 63 |
+
.header-left {
|
| 64 |
+
display: flex;
|
| 65 |
+
flex-direction: column;
|
| 66 |
+
align-items: flex-start;
|
| 67 |
+
gap: 4px;
|
| 68 |
+
margin-left: 8px;
|
| 69 |
+
}
|
| 70 |
+
.header-title {
|
| 71 |
+
font-size: 1.6rem;
|
| 72 |
+
font-weight: 600;
|
| 73 |
+
margin: 0;
|
| 74 |
+
}
|
| 75 |
+
.header-creator {
|
| 76 |
+
font-size: 0.9rem;
|
| 77 |
+
color: var(--text);
|
| 78 |
}
|
|
|
|
| 79 |
|
| 80 |
+
/* Chat scroll area */
|
| 81 |
+
.chat-area {
|
| 82 |
+
height: calc(100vh - 200px);
|
| 83 |
+
overflow-y: auto;
|
| 84 |
+
padding: 1rem 2rem;
|
| 85 |
}
|
| 86 |
|
| 87 |
+
/* Chat bubble styles */
|
| 88 |
.chat-user {
|
| 89 |
background: #2d3748;
|
| 90 |
+
padding: 12px 14px;
|
| 91 |
+
border-radius: 18px 18px 4px 18px;
|
| 92 |
+
margin: 12px 0 12px auto;
|
| 93 |
+
max-width: 75%;
|
| 94 |
+
color: var(--text);
|
| 95 |
+
line-height: 1.4;
|
| 96 |
+
word-break: break-word;
|
| 97 |
}
|
| 98 |
.chat-bot {
|
| 99 |
+
background: var(--primary);
|
| 100 |
+
padding: 12px 14px;
|
| 101 |
+
border-radius: 18px 18px 18px 4px;
|
| 102 |
+
margin: 12px auto 12px 0;
|
| 103 |
+
max-width: 75%;
|
| 104 |
+
color: white;
|
| 105 |
+
line-height: 1.4;
|
| 106 |
+
word-break: break-word;
|
| 107 |
}
|
|
|
|
| 108 |
.sources {
|
| 109 |
+
font-size: 0.8rem;
|
| 110 |
+
opacity: 0.75;
|
| 111 |
+
margin-top: 8px;
|
| 112 |
+
border-top: 1px solid rgba(255,255,255,0.08);
|
| 113 |
+
padding-top: 6px;
|
| 114 |
+
}
|
| 115 |
+
|
| 116 |
+
/* Sticky input bar at bottom of main area */
|
| 117 |
+
.input-bar {
|
| 118 |
+
position: sticky;
|
| 119 |
+
bottom: 0;
|
| 120 |
+
left: 320px;
|
| 121 |
+
right: 0;
|
| 122 |
+
background: var(--bg);
|
| 123 |
+
padding: 12px 20px;
|
| 124 |
+
border-top: 1px solid rgba(255,255,255,0.06);
|
| 125 |
+
z-index: 999;
|
| 126 |
}
|
| 127 |
|
| 128 |
+
/* Form layout */
|
| 129 |
+
.input-row {
|
| 130 |
+
max-width: 980px;
|
| 131 |
+
margin: 0 auto;
|
| 132 |
+
display: flex;
|
| 133 |
+
gap: 8px;
|
| 134 |
+
align-items: center;
|
| 135 |
+
}
|
| 136 |
+
.text-input {
|
| 137 |
+
flex: 1;
|
| 138 |
+
background: transparent;
|
| 139 |
+
border: 1px solid rgba(255,255,255,0.06);
|
| 140 |
+
padding: 10px 14px;
|
| 141 |
+
border-radius: 999px;
|
| 142 |
+
color: var(--text);
|
| 143 |
+
outline: none;
|
| 144 |
+
font-size: 1rem;
|
| 145 |
+
}
|
| 146 |
+
.text-input::placeholder {
|
| 147 |
+
color: rgba(255,255,255,0.45);
|
| 148 |
+
}
|
| 149 |
+
.submit-arrow {
|
| 150 |
+
background: var(--primary);
|
| 151 |
+
color: white;
|
| 152 |
+
border: none;
|
| 153 |
+
height: 42px;
|
| 154 |
+
width: 42px;
|
| 155 |
+
border-radius: 50%;
|
| 156 |
+
display: inline-flex;
|
| 157 |
+
align-items: center;
|
| 158 |
+
justify-content: center;
|
| 159 |
+
font-weight: 700;
|
| 160 |
+
cursor: pointer;
|
| 161 |
+
}
|
| 162 |
+
.submit-arrow:disabled {
|
| 163 |
+
opacity: 0.45;
|
| 164 |
+
cursor: not-allowed;
|
| 165 |
+
}
|
| 166 |
+
|
| 167 |
+
/* Small responsive tweak */
|
| 168 |
+
@media (max-width: 768px) {
|
| 169 |
+
section[data-testid="stSidebar"] {
|
| 170 |
+
position: relative;
|
| 171 |
+
width: 100%;
|
| 172 |
+
height: auto;
|
| 173 |
+
}
|
| 174 |
+
.main {
|
| 175 |
+
margin-left: 0;
|
| 176 |
+
}
|
| 177 |
+
.input-bar { left: 0; }
|
| 178 |
}
|
| 179 |
</style>
|
| 180 |
+
""",
|
| 181 |
+
unsafe_allow_html=True,
|
| 182 |
+
)
|
| 183 |
|
| 184 |
# ---------------- SESSION STATE ----------------
|
| 185 |
if "chat" not in st.session_state:
|
| 186 |
st.session_state.chat = []
|
|
|
|
| 187 |
if "vectorstore" not in st.session_state:
|
| 188 |
st.session_state.vectorstore = None
|
|
|
|
| 189 |
if "retriever" not in st.session_state:
|
| 190 |
st.session_state.retriever = None
|
|
|
|
| 191 |
if "uploaded_file_name" not in st.session_state:
|
| 192 |
st.session_state.uploaded_file_name = None
|
|
|
|
| 193 |
if "uploader_key" not in st.session_state:
|
| 194 |
st.session_state.uploader_key = 0
|
| 195 |
|
| 196 |
+
# ---------------- HELPERS / LOGIC ----------------
|
| 197 |
def clear_chat_history():
|
| 198 |
st.session_state.chat = []
|
| 199 |
|
|
|
|
| 208 |
st.success("Memory cleared. Please upload a new PDF.")
|
| 209 |
|
| 210 |
def process_pdf(uploaded_file):
|
|
|
|
| 211 |
try:
|
| 212 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
|
| 213 |
tmp.write(uploaded_file.getvalue())
|
| 214 |
path = tmp.name
|
| 215 |
+
|
| 216 |
loader = PyPDFLoader(path)
|
| 217 |
docs = loader.load()
|
| 218 |
+
|
| 219 |
+
splitter = RecursiveCharacterTextSplitter(chunk_size=800, chunk_overlap=50)
|
|
|
|
|
|
|
|
|
|
| 220 |
chunks = splitter.split_documents(docs)
|
| 221 |
+
|
| 222 |
embeddings = HuggingFaceEmbeddings(
|
| 223 |
model_name="sentence-transformers/all-MiniLM-L6-v2",
|
| 224 |
model_kwargs={"device": "cpu"},
|
| 225 |
+
encode_kwargs={"normalize_embeddings": True},
|
| 226 |
)
|
| 227 |
+
|
| 228 |
vectorstore = Chroma.from_documents(chunks, embeddings)
|
| 229 |
retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
|
| 230 |
+
|
| 231 |
st.session_state.vectorstore = vectorstore
|
| 232 |
st.session_state.retriever = retriever
|
| 233 |
+
|
| 234 |
if os.path.exists(path):
|
| 235 |
os.unlink(path)
|
| 236 |
+
|
| 237 |
return len(chunks)
|
|
|
|
| 238 |
except Exception as e:
|
| 239 |
+
logging.exception("PDF processing error")
|
| 240 |
+
st.error(f"Error processing PDF: {e}")
|
|
|
|
|
|
|
| 241 |
return None
|
| 242 |
|
| 243 |
def ask_question(question):
|
|
|
|
| 244 |
if not client:
|
| 245 |
return None, 0, "Groq client is not initialized. Check API key setup."
|
|
|
|
| 246 |
if not st.session_state.retriever:
|
| 247 |
return None, 0, "Upload PDF first to initialize the knowledge base."
|
|
|
|
| 248 |
try:
|
| 249 |
docs = st.session_state.retriever.invoke(question)
|
| 250 |
context = "\n\n".join(d.page_content for d in docs)
|
| 251 |
+
|
| 252 |
prompt = f"""
|
| 253 |
You are a strict RAG Q&A assistant.
|
| 254 |
Use ONLY the context provided. If the answer is not found, reply:
|
|
|
|
| 265 |
response = client.chat.completions.create(
|
| 266 |
model=GROQ_MODEL,
|
| 267 |
messages=[
|
| 268 |
+
{"role": "system", "content": "Use only the PDF content. If answer not found, say: 'I cannot find this in the PDF.'"},
|
| 269 |
+
{"role": "user", "content": prompt},
|
|
|
|
| 270 |
],
|
| 271 |
+
temperature=0.0,
|
| 272 |
)
|
|
|
|
| 273 |
answer = response.choices[0].message.content.strip()
|
| 274 |
return answer, len(docs), None
|
|
|
|
| 275 |
except APIError as e:
|
| 276 |
return None, 0, f"Groq API Error: {str(e)}"
|
| 277 |
except Exception as e:
|
| 278 |
+
logging.exception("Ask error")
|
| 279 |
return None, 0, f"General error: {str(e)}"
|
| 280 |
|
| 281 |
+
# ---------------- SIDEBAR (Upload + Controls) ----------------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 282 |
with st.sidebar:
|
| 283 |
+
st.markdown("### Upload PDF", unsafe_allow_html=True)
|
| 284 |
+
uploaded = st.file_uploader("Choose a PDF file", type=["pdf"], key=st.session_state.uploader_key, label_visibility="collapsed")
|
| 285 |
+
if uploaded and uploaded.name != st.session_state.uploaded_file_name:
|
| 286 |
+
# reset chat and process
|
| 287 |
+
st.session_state.chat = []
|
| 288 |
+
st.session_state.uploaded_file_name = None
|
| 289 |
+
with st.spinner(f"Processing '{uploaded.name}'..."):
|
| 290 |
+
chunks_count = process_pdf(uploaded)
|
| 291 |
+
if chunks_count is not None:
|
| 292 |
+
st.success(f"β
PDF processed successfully! {chunks_count} chunks created.")
|
| 293 |
+
st.session_state.uploaded_file_name = uploaded.name
|
| 294 |
+
else:
|
| 295 |
+
st.error("β Failed to process PDF")
|
| 296 |
+
st.session_state.uploaded_file_name = None
|
| 297 |
+
st.experimental_rerun()
|
| 298 |
+
|
| 299 |
+
st.markdown("---")
|
| 300 |
st.header("Controls")
|
| 301 |
st.button("ποΈ Clear Chat History", on_click=clear_chat_history, use_container_width=True)
|
| 302 |
st.button("π₯ Clear PDF Memory", on_click=clear_memory, use_container_width=True)
|
| 303 |
+
|
| 304 |
st.markdown("---")
|
| 305 |
+
st.subheader("Status")
|
| 306 |
if st.session_state.uploaded_file_name:
|
| 307 |
+
st.success(f"β
Active PDF:\n`{st.session_state.uploaded_file_name}`")
|
| 308 |
else:
|
| 309 |
+
st.info("β¬οΈ Upload a PDF to start chatting!")
|
| 310 |
+
|
| 311 |
+
# ---------------- MAIN HEADER (Top-left Title + Creator) ----------------
|
| 312 |
+
# uses "main" margin via CSS
|
| 313 |
+
st.markdown('<div class="main">', unsafe_allow_html=True)
|
| 314 |
+
st.markdown(
|
| 315 |
+
"""
|
| 316 |
+
<div class="header-left">
|
| 317 |
+
<div class="header-title">π PDF Assistant</div>
|
| 318 |
+
<div class="header-creator">Created by <a href="https://www.linkedin.com/in/abhishek-iitr/" target="_blank" style="color: #9fc2ff;">Abhishek Saxena</a></div>
|
| 319 |
+
</div>
|
| 320 |
+
""",
|
| 321 |
+
unsafe_allow_html=True,
|
| 322 |
)
|
| 323 |
|
| 324 |
+
# ---------------- CHAT AREA ----------------
|
| 325 |
+
st.markdown('<div class="chat-area">', unsafe_allow_html=True)
|
| 326 |
+
|
| 327 |
+
if not st.session_state.chat:
|
| 328 |
+
st.markdown('<div style="color:rgba(255,255,255,0.55); padding:20px 0;">Ask a question about your document to start the conversation.</div>', unsafe_allow_html=True)
|
| 329 |
+
else:
|
| 330 |
+
for role, msg in st.session_state.chat:
|
| 331 |
+
if role == "user":
|
| 332 |
+
st.markdown(f"<div class='chat-user'>{msg}</div>", unsafe_allow_html=True)
|
|
|
|
| 333 |
else:
|
| 334 |
+
st.markdown(f"<div class='chat-bot'>{msg}</div>", unsafe_allow_html=True)
|
| 335 |
+
|
| 336 |
+
st.markdown("</div>", unsafe_allow_html=True) # close chat-area
|
| 337 |
+
|
| 338 |
+
# ---------------- INPUT BAR (Enter to submit + arrow button) ----------------
|
| 339 |
+
st.markdown('<div class="input-bar">', unsafe_allow_html=True)
|
| 340 |
+
st.markdown('<div class="input-row">', unsafe_allow_html=True)
|
| 341 |
+
|
| 342 |
+
# Build a form: pressing Enter in the input will submit the form.
|
| 343 |
+
with st.form(key="ask_form", clear_on_submit=True):
|
| 344 |
+
cols = st.columns([1, 0.12])
|
| 345 |
+
with cols[0]:
|
| 346 |
+
q = st.text_input(
|
| 347 |
+
"Type your question",
|
| 348 |
+
key="question_input",
|
| 349 |
+
placeholder="Ask anything about your PDF document...",
|
| 350 |
+
disabled=(st.session_state.uploaded_file_name is None or client is None),
|
| 351 |
+
label_visibility="collapsed",
|
| 352 |
+
)
|
| 353 |
+
with cols[1]:
|
| 354 |
+
# Arrow submit button (visible). Pressing Enter also triggers the form submit.
|
| 355 |
+
submit = st.form_submit_button("β€", help="Send (Enter or click)", disabled=(st.session_state.uploaded_file_name is None or client is None))
|
| 356 |
+
|
| 357 |
+
st.markdown("</div></div>", unsafe_allow_html=True) # close input-row and input-bar
|
| 358 |
+
|
| 359 |
+
if submit and q:
|
| 360 |
+
# Append user message
|
| 361 |
+
st.session_state.chat.append(("user", q))
|
| 362 |
+
|
| 363 |
+
with st.spinner("Thinking..."):
|
| 364 |
+
answer, sources, error = ask_question(q)
|
| 365 |
+
if answer:
|
| 366 |
+
bot_msg = f"{answer}<div class='sources'>Context Chunks Used: {sources}</div>"
|
| 367 |
+
st.session_state.chat.append(("bot", bot_msg))
|
| 368 |
+
else:
|
| 369 |
+
st.session_state.chat.append(("bot", f"π΄ **Error:** {error}"))
|
| 370 |
+
|
| 371 |
+
# Rerun so new messages show and form clears
|
| 372 |
+
st.experimental_rerun()
|