app.py

import streamlit as st
import google.generativeai as genai
import numpy as np

# Configure Gemini API
genai.configure(api_key=st.secrets["GEMINI_API_KEY"])

st.title("Text Embedding Similarity Test")

def split_into_chunks(text, chunk_size=500):
    """Split text into chunks of approximately specified character length"""
    return [text[i:i+chunk_size] for i in range(0, len(text), chunk_size)]

def get_embedding(text):
    """Get embedding for a single text chunk"""
    return genai.embed_content(
        model="models/text-embedding-004",
        content=text
    )['embedding']

def cosine_similarity(vec1, vec2):
    """Compute cosine similarity between two vectors"""
    return np.dot(vec1, vec2) / (np.linalg.norm(vec1) * np.linalg.norm(vec2))

# Text input areas
col1, col2 = st.columns(2)
with col1:
    input_text1 = st.text_area("Enter your first text:", 
                             height=200,
                             placeholder="Type or paste your first text here...")

with col2:
    input_text2 = st.text_area("Enter text to compare:", 
                             height=200,
                             placeholder="Type or paste text to compare...")

if st.button("Run Similarity Test"):
    if not input_text1.strip() or not input_text2.strip():
        st.warning("Please enter text in both input fields.")
    else:
        with st.spinner("Analyzing texts..."):
            try:
                # Process first text into chunks
                chunks = split_into_chunks(input_text1)
                if len(chunks) > 1:
                    st.info(f"Split first text into {len(chunks)} chunks")
                
                # Generate embeddings for all chunks
                embeddings = [get_embedding(chunk) for chunk in chunks]
                
                # Generate embedding for comparison text
                compare_embedding = get_embedding(input_text2)
                
                # Calculate similarities
                similarities = [cosine_similarity(emb, compare_embedding) for emb in embeddings]
                max_score = max(similarities)
                max_index = similarities.index(max_score)
                
                # Display results
                st.subheader("📊 Similarity Results")
                st.write(f"**Highest similarity score:** {max_score:.4f}")
                
                st.subheader("🧩 Most Similar Chunk")
                st.write(chunks[max_index])
                
                st.subheader("📈 All Chunk Similarities")
                for i, (chunk, score) in enumerate(zip(chunks, similarities)):
                    st.write(f"Chunk {i+1} ({len(chunk)} chars): {score:.4f}")
                    st.expander(f"View chunk {i+1}").write(chunk)
                
            except Exception as e:
                st.error(f"Error processing texts: {str(e)}")