Meet ShopSense
Track shipments in real time, get instant AI-powered refund decisions, and discover products that actually match your taste, not just "people also bought" noise.
Live Tracking
Unified carrier API aggregation, FedEx, UPS, USPS, DHL normalized into a single real-time timeline with proactive delay alerts
AI Refund Agent
Understands return policies, product condition, and order history, resolves 80% of refund requests in under 8 minutes
Taste Profiling
Builds preference vectors from browsing, purchases, and reviews, then surfaces products with explainable match scores
Happy Customers
Post-purchase experience that actually works, real-time updates, instant resolutions, and recommendations that land
Commerce Intelligence Graph
Watch customer signals flow through AI processing layers in real time. Hover nodes to see the data transform.
Signal: Interest vector [0.8, 0.3, 0.6]
Signal: Purchase intent 0.72
Signal: Satisfaction delta −0.35
Signal: Sentiment +0.82 (audio quality)
Updated every 50ms during session
Model: all-MiniLM-L6-v2
Classifier
Track | Refund | Recommend | Support
Confidence: 0.94
RAG
Top-k=5, reranked by recency
Store: Pinecone (1536 dims)
Engine
Category affinity + price sensitivity
Output: Ranked product list
Agent
Auto-approve | Manual | Deny
SLA: < 8 minutes
Intel
Proactive delay detection (24h window)
Output: Timeline + ETA
94% match, Sony WH-1000XM5
91% match, Keychron Q1 Pro
Card credited in 2-3 business days
Confidence: 0.97
Proactive SMS + email sent
Compensation coupon auto-generated
A/B test: +12% CTR on recs
Drift detection: monthly recalibration
Your Orders, Unboxed
Real-time shipment tracking, instant refund status, and personalized recommendations, all in one place.
Hey there! I'm ShopSense, your personal shopping assistant. I can track your orders, handle refunds instantly, and recommend products you'll actually love. What can I do for you?
Commerce Intelligence
The Challenge
BrightCart was losing customers to poor post-purchase experience. Shipment tracking was a black hole, customers got a tracking number and nothing else until delivery (or a missed delivery). Refund requests took 5-7 business days because every case went through manual review. And their recommendation engine was the usual "frequently bought together" algorithm that suggested phone cases to someone who bought a laptop. Customers were frustrated, support tickets were piling up at 400+ per day, and repeat purchase rate had dropped to 18%.
The Approach
I designed a three-layer intelligence system. Layer 1: a real-time shipment tracking engine that aggregates carrier APIs (FedEx, UPS, USPS, DHL), normalizes status events into a unified timeline, and sends proactive notifications when delays are detected, before the customer even asks. Layer 2: an AI-powered refund agent built on Claude API that understands return policies, product condition context, and order history to resolve 80% of refund requests instantly with fair, consistent decisions. Layer 3: a taste-profiling recommendation engine that builds customer preference vectors from browsing patterns, purchase history, review sentiment, and explicit preferences, then uses cosine similarity to surface genuinely relevant products with explainable match scores.