Tagline: A context-aware AI shopping assistant that blends user memory, store data, and real-time environmental signals to deliver relevant, human-like retail guidance — instantly.
👉 https://huggingface.co/spaces/Sagar8528/Context_Retail_AI
This is a working proof-of-concept demonstrating the core intelligence engine. Not production-grade yet — but the foundation is strong.
Physical retail still treats every customer like a stranger.
Walk into any Starbucks, Domino’s, or Decathlon and you'll see:
- No personalization
- No recollection of past behavior
- No situational awareness (weather, distance, offers, etc.)
- No predictive intent
Meanwhile, online commerce knows:
- What you like
- What you bought
- What you’re likely to want next
Offline retail is missing this intelligence layer.
Digital personalization exists. Physical personalization doesn’t.
A context-aware retail assistant that:
✔ remembers the customer ✔ understands intent and emotion ✔ reacts to real-world conditions (weather, location) ✔ recommends items intelligently ✔ nudges purchases like a trained human associate — at scale
No forms. No onboarding. No typing preferences.
Just:
User: i'm thirsty Bot → "Pune is hot today (29°C). There's a Starbucks 1.2km away — an iced caramel latte would be perfect."
- Upload a CSV once (customer list, products, offers)
- System auto-learns preferences over time
- No additional training needed
The system uses three categories of signals:
| Type | Example | Source |
|---|---|---|
| Static | Offers, menu, store metadata | Merchant CSV |
| Dynamic User | Preferences, patterns, sentiment | Memory + Chat |
| Real-Time Context | Weather, distance, availability | External APIs |
User Message ↓ Intent Detection ↓ Context Fusion ├─ User memory (dynamic) ├─ Store DB (static) ├─ Location + Weather (external dynamic) ↓ Decision Engine ↓ LLM Response Generation (Groq) ↓ Memory Update (if new useful info) ↓ Reply to User
Detects expressions like "cold", "thirsty", "nearby", "sweet", "recommend".
Stores only useful signals:
"I love cold brew" → stored
"Tell me a joke" → ignored
Uses:
- IP → City (fallback: store default)
- Weather API → temperature + condition
- Store locator → nearest store
Creates a scored recommendation based on:
- Weather influence
- Loyalty tier
- Customer preferences
- Distance from store
- Relevant seasonal items
Groq (Llama3) generates human-like replies, not robotic text.
| Layer | Technology |
|---|---|
| Backend | Python |
| AI Model | Groq (Llama-3.1-8B-Instant) |
| Memory DB | MongoDB Atlas |
| Weather | OpenWeather API |
| Location | IP-based geo lookup (demo) |
| UI | Gradio |
| Deployment | Hugging Face Spaces |
| Area | Current State | Goal |
|---|---|---|
| Authentication | User enters name manually | JWT login + persistent profile |
| Location Accuracy | IP detection + text extraction | GPS permissions + reverse geolocation |
| Weather Logic | Single API | Retry layer + offline fallback |
| Recommendation Engine | Rule-based + LLM | Hybrid: vector retrieval + ranking model |
| Product Catalog | CSV only | Dynamic update + vendor dashboard |
📌 Prototype is functional. 📌 Real-time reasoning works. 📌 Memory retention works. 📌 Weather & location influence responses.
Next version will include:
- UI polish
- Product embedding-based RAG
- Conversion tracking (analytics)
This isn’t just a chatbot — it’s a personalized bridge between AI and physical retail, meant to increase repeat visits, conversion rates, and customer loyalty at scale.
One assistant → millions of personalized retail experiences.