✨ Features

This section highlights Snapr’s core features and agents — from category selection to finding the best-value deals through smart discovery and filtering.

Click on image to zoom

🧠 Planning Agent

  • Orchestrates the full pipeline:
    • Scans deals via DealScannerAgent,
    • Predicts prices using multiple agents (FTPriceAgent, XGBoostPriceAgent, RAGPriceAgent, EnsemblePriceAgent)
    • Filters deals based on a fixed discount threshold
    • And saves accepted deals to shared memory.
  • Logs each decision for transparency.
  • Produces a structured summary of accepted deals with price, estimated value, discount, and link.

🟢 Deal Scanner Agent

  • Fetches real-time deals from RSS feeds based on selected categories.
  • Already-seen deals are automatically skipped using a memory system.
  • OpenAI (GPT) selects the top 5 deals by filtering for clear descriptions and exact numeric prices.
  • Results are returned in a consistent, machine-readable JSON format.
  • Users can view real-time logs of what the agent is doing (e.g., fetched deals, skipped, filtered).

🔴 Fine-Tuned LLM Agent

  • Uses a fine-tuned (FT) LLaMA model to predict prices from product descriptions.
  • Runs remotely on Modal for scalable execution.
  • Provides logs on remote calls and failures.

🔵 RAG-Based Agent

  • Predicts prices using a Retrieval-Augmented Generation pipeline.
  • Combines E5 embeddings, ChromaDB, and a frontier LLM (OpenAI GPT).
  • Executes remotely via Modal and logs embedding/model activity.

🟡 XGBoost-Based Agent

  • Uses E5 embeddings + XGBoost for regression-based price prediction.
  • Runs remotely on Modal infrastructure.
  • Logs model loading and predictions visibly.

🟣 Ensemble Agent

  • Combines predictions from FT, RAG, and XGBoost agents.
  • Applies a trained linear regression model to produce a final estimated price.
  • Logs individual predictions and final computed result.