resume-screening-rag / src /streamlit_app.py
CPSC481
Fix Tab 1: replace PDF URL input with drag-and-drop file uploader
f37711c
Raw History Blame Contribute Delete
4.31 kB
from __future__ import annotations
import streamlit as st
import config
from embeddings import embed_texts, embed_query, embedding_dim
from vector_store import PineconeVectorStore
from ingest import build_vectors_for_pdf
from search import search_resumes
NAMESPACE = "resumes"
st.set_page_config(page_title="Resume Screening (RAG)", layout="wide")
st.title("Resume Screening System (RAG)")
@st.cache_resource
def get_vector_store():
dim = embedding_dim(config.EMBED_MODEL_NAME)
return PineconeVectorStore(
api_key=config.PINECONE_API_KEY,
index_name=config.PINECONE_INDEX_NAME,
dimension=dim,
cloud=config.PINECONE_CLOUD,
region=config.PINECONE_REGION,
)
vs = get_vector_store()
tab1, tab2, tab3 = st.tabs(["Upload & Ingest", "Search", "Admin"])
with tab1:
st.subheader("Upload & Ingest PDF Resumes")
uploaded_file = st.file_uploader("Upload a PDF resume", type=["pdf"])
colA, colB, colC = st.columns(3)
with colA:
chunk_size = st.number_input("Chunk size (chars)", min_value=200, max_value=4000, value=config.CHUNK_SIZE, step=100)
with colB:
overlap = st.number_input("Overlap (chars)", min_value=0, max_value=1000, value=config.CHUNK_OVERLAP, step=50)
with colC:
namespace = st.text_input("Namespace", value=NAMESPACE)
if st.button("Ingest", type="primary", disabled=not uploaded_file):
try:
with st.spinner("Processing and uploading to Pinecone..."):
vectors = build_vectors_for_pdf(
pdf_bytes=uploaded_file.read(),
filename=uploaded_file.name,
embed_fn=lambda texts: embed_texts(texts, model_name=config.EMBED_MODEL_NAME),
chunk_size=int(chunk_size),
overlap=int(overlap),
)
vs.upsert(vectors=vectors, namespace=namespace)
st.success(f"Done. Upserted {len(vectors)} chunks into namespace '{namespace}'.")
except Exception as e:
st.error(f"Ingest failed: {e}")
with tab2:
st.subheader("Search candidates")
namespace = st.text_input("Namespace to search", value=NAMESPACE, key="search_namespace")
query = st.text_input("Enter a keyword or query", placeholder="e.g., Python SQL ETL 3 years, machine learning internship")
col1, col2 = st.columns([1, 1])
with col1:
top_k = st.slider("Top K results", min_value=3, max_value=20, value=config.TOP_K, step=1)
with col2:
min_score = st.slider("Minimum similarity score", min_value=0.0, max_value=1.0, value=0.2, step=0.05)
if st.button("Search", type="primary", disabled=not query.strip()):
try:
with st.spinner("Searching..."):
results = search_resumes(
query=query.strip(),
embed_query_fn=lambda q: embed_query(q, model_name=config.EMBED_MODEL_NAME),
vector_store=vs,
top_k=int(top_k),
namespace=namespace,
)
results = [r for r in results if r["score"] >= float(min_score)]
if not results:
st.warning("No matches found (try lowering minimum score or ingest more resumes).")
else:
st.write(f"Found {len(results)} matches:")
for r in results:
with st.container(border=True):
st.markdown(
f"**{r['filename']}** \n"
f"Score: `{r['score']:.3f}` | Page: `{r['page']}` | Chunk: `{r['chunk_index']}`"
)
st.write(r["text"][:1200] + ("..." if len(r["text"]) > 1200 else ""))
except Exception as e:
st.error(f"Search failed: {e}")
with tab3:
st.subheader("Admin")
namespace = st.text_input("Namespace to clear", value=NAMESPACE, key="admin_namespace")
st.warning("This will delete ALL vectors in the namespace.")
if st.button("Delete namespace vectors", type="secondary"):
try:
vs.delete_namespace(namespace=namespace)
st.success(f"Deleted all vectors in namespace '{namespace}'.")
except Exception as e:
st.error(f"Delete failed: {e}")