Behavior Prediction with RAG + Vectors — A minimal, extensible system that retrieves similar past episodes with FAISS/Chroma and predicts a user’s next move & mood. Shows model-architecture thinking, vector DB usage, and prompt-based personalization (PyTorch/TensorFlow-ready).
Behavior AI — Retrieval-Augmented Human Behavior Prediction
This project is a compact playground for predicting/mock-predicting human decisions (e.g., next move, mood, suggested response). It combines semantic embeddings + vector search (FAISS/Chroma) with a simple baseline (k-NN majority vote) and is structured to plug in a PyTorch/TensorFlow head later. You’ll see end-to-end model architecture thinking, vector DB usage, and prompt engineering for personalized outputs.
Why this repo?
-Personalization via retrieval: Find similar past “episodes” and ground predictions in real examples.
-Deep ML-ready: Start with k-NN; upgrade to a small PyTorch/TensorFlow classifier anytime.
-Vector DB skills: FAISS (or Chroma) indexing, querying, and metadata-aware search.
-Prompted summaries: Optional natural-language responses that cite nearest neighbors.
Tech Stack
-Embeddings: Sentence-Transformers
-Vector DB: FAISS (default) or ChromaDB
-ML: PyTorch/TensorFlow (pluggable)
-Utilities: Pydantic, NumPy, scikit-learn