SmartNeed

Inspiration

Most eCommerce platforms are built to sell, not to understand. We noticed a common pain point—people often don’t get what they truly need, but what the system wants to push. That insight sparked the idea for SmartNeed, an engine that uses natural language understanding to match user intent with the right products.

What We Learned

1. How to use LLMs to translate natural language into actionable search filters
2. Importance of clean product data and structured tagging for better retrieval
3. Techniques for combining classification, vector search, and retrieval-augmented generation (RAG)

How We Built It

1. FastAPI backend to serve APIs and handle query processing
2. MongoDB Atlas for storing and searching the product catalog
3. LLMs (Gemini) to understand user intent and recommend products
4. React frontend that lets users describe what they’re looking for in plain language
5. Semantic search using embeddings and prompt-based filtering for accurate results

Challenges We Faced

1. Interpreting vague or complex user queries into structured filters
2. Ensuring real-time performance while calling LLM APIs
3. Balancing cost, speed, and accuracy at scale

Built With

• fastapi
• javascript
• mongodb
• python
• react