PERSONAL PROJECT · PRODUCT Case Study
PriceWatch: AI-Shopping Assistant
Live Demo
Product Strategy
PRD & Scoping
AI Matching
System Architecture
Prioritization
Web Scraping
Notification Design
Role
Solo Product Owner: end-to-end scoping, PRD Writing, prioritisation & hands on build
Focus
Scoping a real life price-tracking problem into a buildable MVP backed by a documented tradeoff
Outcome
A MVP with a live demo, a documented PRD, and a decision trail from problem to build
A price-tracking assistant that watches Amazon and Flipkart for considered purchase items and alerts you when a real price drop happens - built end to end, from PRD to backend, as a hands-on PM exercise.
Role: Solo Product Owner and builder - scoping, PRD, prioritization, documentation and implementaton
Focus: Turning a personal frustration (missing genuine price drops during sale season) into a scoped, buildable product
Outcome: A working MVP with a live demo, a documented PRD and a decision train showing how the scope was cut
The Problem
A user spends their time looking for price drops across multiple e-commerce sites with the hope of getting a discount. However, manually watching prices across retailers and sale window is tedious, and most existing tools track only MRP and do not provide a side-by-side comparision of prices across different platforms
The Goal
A tool that tracks the actual selling price over time, matches the same product across retailers and only notifies when the price genuinely changes
Scoping it as a PM would
Decision
Why
Amazon + Flipkart only, electronics only
Narrow surface area to prove the mechanism before generalizing to categories with messier attribute matching (fashion sizing, grocery units)
Human-confirmed cross-retailer matching, not silent auto-match
An AI-only match risks a false-positive merge — showing “price drop” on the wrong product. A human-in-the-loop step protects trust in every alert
Re-verify AI matches only when scraped title text changes
Re-running AI verification on every poll wastes cost and latency for a signal that rarely changes
Price baseline = actual shown selling price, not claimed MRP
Directly kills the fake-discount problem the tool exists to solve — no MRP-based discount math anywhere in the product
Historical low is a one-time, cross-checked lookup, shown separately from the live trend line
Keeps the tracked trend graph trustworthy and auditable, without pretending pre-tracking context is live data
Twice-daily polling, with 30-60 min bursts during known sale windows
Matches polling cost to when prices actually move, instead of polling uniformly and missing fast sale-window changes
Guest/demo account alongside the real login
Lets anyone — including a recruiter — try the product with sample data, zero setup friction
System overview
The pipeline: a product is added via search, an AI step extracts attributes and proposes a cross-retailer match which a human confirms, a scheduled poller checks prices at a cadence tuned to sale activity, results land in a price database that separates the live trend from the one-time historical-low lookup, and genuine drops trigger push and email notifications plus an in-app alert history.
What's next
Including multiple e-commerce retailers for a comprehensive price tracking experience
User Profile that displays overall amount saved through PriceWatch per user
Suggested search based on user history
AI-Assisted shopping decision maker
Smarter burst-window detection instead of hardcoded sale dates
Full PRD available on request