github-doc-rag / app.py
ashiquzzaman's picture
Update app.py
f16672c verified
Raw History Blame Contribute Delete
3.81 kB
import os
import streamlit as st
from langchain_community.document_loaders import DirectoryLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_pinecone import PineconeVectorStore
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_groq import ChatGroq
from langchain_core.prompts import ChatPromptTemplate
from langchain.chains import create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
import pinecone
# πŸ”‘ ENV VARS
PINECONE_API = os.getenv("PINECONE_API_KEY")
GROQ_API = os.getenv("GROQ_API_KEY")
INDEX_NAME = "github-docs-rag"
DOCS_PATH = "./target_docs"
@st.cache_resource
def build_rag_chain():
pc = pinecone.Pinecone(api_key=PINECONE_API)
embedder = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
existing_indexes = pc.list_indexes().names()
if INDEX_NAME not in existing_indexes:
with st.spinner("πŸ“₯ Loading docs & building vector index..."):
pc.create_index(
name=INDEX_NAME,
dimension=384,
metric="cosine",
spec=pinecone.ServerlessSpec(cloud="aws", region="us-east-1")
)
loader = DirectoryLoader(DOCS_PATH, glob="**/*.md")
docs = loader.load()
if not docs:
st.error("❌ No documents found! Check target_docs/ folder.")
st.stop()
splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
chunks = splitter.split_documents(docs)
PineconeVectorStore.from_documents(chunks, embedder, index_name=INDEX_NAME)
st.success(f"βœ… Indexed {len(chunks)} chunks from {len(docs)} files.")
else:
st.info("πŸ“‚ Vector index loaded.")
index = pc.Index(INDEX_NAME)
# πŸ”§ FIX: Increase k to get more results
retriever = PineconeVectorStore(index=index, embedding=embedder).as_retriever(
search_kwargs={"k": 5} # Changed from 3 to 5
)
llm = ChatGroq(model="llama-3.1-8b-instant", temperature=0.3, api_key=GROQ_API)
# πŸ”§ FIX: More flexible prompt that actually uses the context
prompt = ChatPromptTemplate.from_template(
"""You are a helpful documentation assistant. Use the provided context to answer the question.
Context:
{context}
Question: {input}
Answer:"""
)
stuff_chain = create_stuff_documents_chain(llm, prompt)
return create_retrieval_chain(retriever, stuff_chain)
# πŸ–₯️ UI
st.set_page_config(page_title="GitHub Docs RAG", layout="wide")
st.title("πŸ“¦ GitHub Docs Search")
if not PINECONE_API or not GROQ_API:
st.error("⚠️ Missing API keys in Settings.")
st.stop()
rag_chain = build_rag_chain()
# πŸ” Debug: Show how many docs are indexed
with st.expander("πŸ” Debug Info"):
pc = pinecone.Pinecone(api_key=PINECONE_API)
index = pc.Index(INDEX_NAME)
stats = index.describe_index_stats()
st.write(f"**Total vectors in index:** {stats.get('total_vector_count', 0)}")
st.write(f"**Dimensions:** {stats.get('dimension', 384)}")
query = st.text_input("Ask about the docs:", placeholder="e.g., How do I install?")
if query:
with st.spinner("πŸ” Searching..."):
result = rag_chain.invoke({"input": query})
# Show answer
st.subheader("βœ… Answer")
st.write(result["answer"])
# πŸ”§ FIX: Always show retrieved context
st.divider()
st.subheader("πŸ“– Retrieved Context (Top 5)")
for i, doc in enumerate(result["context"], 1):
with st.expander(f"Chunk {i} - Source: `{doc.metadata.get('source', 'unknown')}`"):
st.text(doc.page_content)