MissionCart
Goal-to-cart occasion OS for Amazon Now with 6 working features
Shopping by intent, not keyword
Amazon Now's search is excellent for a specific product you already know you want. Planning a birthday party, setting up a home office, or preparing for a festival involves 15+ products across categories — none of which you're searching for by name. Team B320 built MissionCart to bridge this gap at Amazon HackOn 2026: one sentence in, full cart out.
Context
- 48-hour hackathon sprint
- Must integrate with Amazon Now's product catalog semantically
- React Native frontend + Python FastAPI backend
- Team of 4 across product, ML, and mobile
Our approach
Key decisions
Groq LLaMA 3 for occasion parsing
Groq's inference speed was critical in a 48-hour sprint. LLaMA 3 parses an occasion into structured slot fills: event type, guest count, dietary preferences, and budget ceiling.
FAISS for product retrieval
Embedded the Amazon Now product catalog, retrieving top candidates per occasion slot. Semantic search outperformed keyword lookup — "candles" → "birthday décor" without manual mapping.
BLaIR for cart ranking
Bi-encoder re-ranking filtered FAISS candidates to the final cart, balancing relevance, availability, and price ceiling — all in a single re-rank pass.
6 features in 48 hours
Occasion cart, budget guardrails, dietary filters, substitution suggestions, delivery time estimates, one-tap reorder. Scope was fixed; execution was the variable.
Results
What we achieved
Features shipped in 48 hours
Cart generation from text prompt
AI models in pipeline
Amazon 2026 finalist
Stack used
Next project
Varun Vastralaya Automation →Start a project
Let's build something great
Drop your details below. We'll have it open in Gmail, ready to send — no copy-pasting.