Technical Documentation

1. Project Overview

ElectroFind is a final-year B.E. Artificial Intelligence project that solves the electronics price comparison problem using a combination of offline AI-trained models and real-time web data aggregation. The system has two core layers:

2. Data Pipeline

The offline data pipeline was built to collect, clean, and transform a large dataset of electronics products for model training.

# Simplified data pipeline (Python)
import pandas as pd
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.preprocessing import MinMaxScaler

# Load raw scraped data
df = pd.read_csv('raw_electronics.csv')

# Clean titles
df['clean_title'] = df['title'].str.lower()
  .str.replace(r'[^a-z0-9 ]', '', regex=True)
  .str.strip()

# Remove duplicates based on title similarity
df = df.drop_duplicates(subset=['clean_title', 'platform'])

# Normalize prices
df['price_usd'] = df['price'].apply(normalize_price)

# TF-IDF vectorization
vectorizer = TfidfVectorizer(max_features=5000, ngram_range=(1,2))
tfidf_matrix = vectorizer.fit_transform(df['clean_title'])

3. AI / ML Model

The AI model serves two purposes: cross-platform product matching (deduplication) and validation of the Buy Score signal weights.

# Cosine similarity matching
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np

def find_matches(query_title, product_matrix, vectorizer, threshold=0.85):
    query_vec = vectorizer.transform([query_title])
    similarities = cosine_similarity(query_vec, product_matrix).flatten()
    matches = np.where(similarities >= threshold)[0]
    return matches, similarities[matches]

# Buy Score calculation
def buy_score(price, rating, reviews, min_p, max_p):
    r_score  = (rating / 5.0) * 40
    rv_score = min(math.log10(reviews + 1) / math.log10(100_000) * 30, 30)
    p_score  = (1 - (price - min_p) / (max_p - min_p)) * 30 if max_p > min_p else 15
    return round(r_score + rv_score + p_score, 1)

4. System Architecture

// Simplified API route flow
export async function GET(req: NextRequest) {
  const q = req.nextUrl.searchParams.get("q");

  // 1. Fetch all platforms in parallel (each has own 40s timeout)
  const [oxyResult, serperResult] = await Promise.all([
    fetchAllPlatforms(q),    // Amazon + Flipkart + Best Buy + eBay
    fetchSerperShopping(q),  // Google Shopping (with images)
  ]);

  // 2. Merge, score and sort
  const all    = [...oxyResult.products, ...serperResult.products];
  const scored = calculateScores(all).sort((a, b) => b.score - a.score);

  return NextResponse.json({ products: scored, errors, total: scored.length });
}

5. Buy Score Algorithm

Every product receives a Buy Score (0–100) calculated from three signals validated against our training dataset.

6. API Integrations

7. Search Flow