AI Project
Vibes Recipe - AI Recipe Recommendation App
A semantic search and LLM-powered recommendation system that finds recipes based on what you are in the mood for, not just what's in your pantry.
What Was Built
Built a RAG-based recipe recommendation system that understands natural language queries to surface recipes matching a specific mood or vibe. The flow combines data ingestion, embedding generation, vector similarity search, and LLM-powered response generation to deliver recommendations that feel relevant and fast.
Result
Improved recommendation relevance by 75% while keeping query latency under 3 seconds, turning an open-ended "what should I cook" question into a focused, personalized answer.
Tech Stack
PythonFastAPISentence TransformerspgvectorPostgreSQLRAG Pipeline
AI Models Used
Sentence-Transformers (384-dim Embeddings)Groq LLMCerebras LLMClaude Code (Opus 4.6)
Things Learned
How to build embedding pipelines that scale across large, messy real-world datasets.
How post-retrieval filtering meaningfully improves the quality of LLM-generated recommendations.
How choosing the right vector search strategy directly impacts both relevance and response speed.