AI systemFull-Stack AI Engineer
ReviewLens
Overview
Online shoppers rarely read more than a few reviews, and star ratings alone don't show what's actually good or bad about a product. ReviewLens fixes that for Daraz.pk. Paste a product link and it collects every review, works out which features matter for that product, and shows how buyers rate each one, like Camera 4.6, Battery 3.1, Value for Money 4.2.
Behind the scenes, a Playwright scraper pages through all the reviews. pandas then cleans them and removes duplicates. A Groq LLM generates product-specific categories, and my JEV router, built with LangGraph, sorts each review into those categories with its own rating and sentiment. A review like "great camera but slow processor" counts toward both features, and general praise goes to Overall. Reviews are processed in parallel batches, each response is checked against strict Pydantic schemas and broken JSON gets repaired automatically. The results are pulled together into a Next.js dashboard.
The result: hundreds of reviews summed up feature by feature in seconds, so buyers can see what's good about a product and what isn't before they buy.
What I built
- 01
Full-review scraper
A Playwright scraper that checks the Daraz link and pages through every review, not just the first few.
- 02
Product-aware categories
A Groq LLM picks the features worth rating for each product, so a phone gets Camera and Battery while a kettle gets Heating and Build.
- 03
JEV review router
A LangGraph pipeline that maps each review to the features it mentions and rates each one; general praise goes to Overall.
- 04
Parallel batch processing
Reviews are batched and sent to Groq at the same time, with Pydantic validation and automatic repair of broken JSON.
- 05
pandas EDA pipeline
Cleans and normalizes reviews, then works out how many people mention each feature, the average rating and the sentiment split, saving each stage as CSV.
- 06
Feature-level dashboard
A Next.js and Tailwind UI showing a rating for each feature, a sentiment breakdown and real review quotes.