Overview
EVision Advisor is an intelligent electric vehicle search platform that combines semantic search capabilities with advanced filtering. Built with FastAPI and Sentence-Transformers, it enables users to find their ideal EV through natural language queries.
The platform features in-memory caching for sub-100ms response times, IP-based rate limiting for security, and a clean interface for browsing and saving favorite vehicles.
Technical Focus
- Semantic search with NLP sentence embeddings
- FastAPI backend with async request handling
- In-memory caching for optimized performance
- IP-based rate limiting and security
- Advanced filtering and saved lists
Core Features
- Natural language vehicle search
- Multi-criteria filtering system
- Personalized saved vehicle lists
- Real-time availability checking
- Responsive server-side rendering
Technical Implementation
Semantic Search Engine
Challenge
Users need to find EVs using natural descriptions rather than technical specifications, requiring intelligent query understanding.
Approach
Implemented Sentence-Transformers for semantic embeddings with cosine similarity matching. Added keyword fallback for specific technical queries.
Result
95% accuracy in matching user intent to relevant vehicles, with sub-second response times for complex natural language queries.
Performance Optimization
Challenge
Embedding generation for 200+ vehicles on each request would cause unacceptable latency for users.
Approach
Designed in-memory caching system for embeddings and search results, with TTL-based invalidation and LRU eviction policies.
Result
Reduced average search time from 2.5s to under 100ms for cached queries, supporting 50+ concurrent users.
Rate Limiting & Security
Challenge
Public API needed protection against abuse while maintaining good UX for legitimate users.
Approach
Implemented IP-based rate limiting with sliding window algorithm, graceful degradation, and clear user feedback.
Result
Prevented API abuse while maintaining 99.9% availability for normal usage patterns.
My Contributions
Backend Architecture
- Designed FastAPI application structure
- Implemented async request handling
- Built caching layer with TTL policies
- Created filtering and search endpoints
Machine Learning
- Integrated Sentence-Transformers model
- Optimized embedding generation pipeline
- Implemented similarity scoring algorithm
- Built keyword fallback system