Overview

SaleSniffer checks whether online sale prices are genuine. It tracks products from Sapphire, a Pakistani fashion retailer, records every price it sees over time, and uses that history to judge whether a discount is a real price drop or a markup followed by a markdown. Shoppers can browse verdicts or paste a product link to check it.

My role

This was a two-person final-year project. My teammate wrote the original Sapphire scraper and the rule-based verification engine with its unit tests. I built the Next.js frontend end to end, added the FastAPI service that exposes products, verdicts and admin jobs, extended the scraper with single-product lookups and image handling, wrote the price-history seeding and data clean-up scripts, and merged both codebases into one monorepo.

What it does

• Scrapes Sapphire's sale listing pages and stores each product with an append-only price history.
• Labels every product REAL, FAKE or UNVERIFIED with a confidence score and plain-language reasons.
• Lets visitors filter by verdict and brand, search, sort, and open a price-history chart for any product.
• Resolves a pasted product URL: if the item isn't tracked yet, it is scraped, stored and analysed on demand.
• Gives admins a password-protected panel to run scrapes, re-run verification and seed demo data, with live progress.

How it's built

The backend is Python: requests and BeautifulSoup parse the server-rendered listing pages, PyMongo writes to MongoDB, and FastAPI serves a paginated, filterable REST API. The verification engine turns each price history into features (average, minimum, maximum, standard deviation, number of price changes, a recent-spike check and the real discount against the original price), weighs real against fake signals, and declines to judge when there are fewer than three data points. The frontend is Next.js 16 with React 19, Tailwind CSS 4, SWR for data fetching, Recharts for price charts and GSAP for landing-page animation.

Highlights

• Atomic price tracking: each scraped product is saved with one find-and-update upsert that sets current fields, pushes a timestamped price and reports whether it was inserted, updated or changed price. Images are only overwritten when a real photo is found, so they never regress.
• Background admin jobs: scrape, analyse and seed tasks run in a background thread behind a lock; a second job is rejected with HTTP 409, and the admin page polls status every 1.5 seconds only while a job is running.
• Clearable demo data: since real price history takes weeks to build, an admin action generates flagged synthetic histories for demos, which a later real scrape can wipe.