Deal Hunter

A voice-first shopping copilot. Ask for a product out loud and Deal Hunter searches live web data for current listings, ranks them cheapest-first, reads the best options back to you, and remembers items you want to watch.

Built on three technologies, one job each:

• LiveKit — real-time voice (speech-to-text, the LLM brain, and text-to-speech) via LiveKit Inference.
• Bright Data — live web search (SERP + Web Unlocker) for real, current product prices.
• Moss — semantic retrieval and per-user memory (your watch list, budget, and preferences).

How it works

You speak
  -> LiveKit STT (speech -> text)
  -> LLM decides to call a tool
       -> find_deals      -> Bright Data SERP -> live listings -> ranked cheapest-first
       -> remember_fact   -> Moss memory (watch this / my budget)
       -> recall_facts    -> Moss memory (scoped to you)
  -> LiveKit TTS (spoken summary)  +  live "Best Prices" cards in the browser

Project structure

agent-py/                 # Python LiveKit voice agent
  src/agent.py            # agent entrypoint + tools (find_deals, memory, knowledge)
  src/deal_hunter/
    finder.py             # price search: Bright Data SERP -> ranked deals -> voice summary
    web_research.py       # Bright Data SERP + Unlocker helpers
  src/create_index.py     # seeds the Moss knowledge + memory indexes
  tests/                  # pytest suite
frontend/                 # Next.js app: landing + live voice UI with deal cards

Prerequisites

• Python 3.10+ and uv
• Node.js 22+ and pnpm 10+
• A LiveKit Cloud project
• Moss project credentials
• A Bright Data account with a SERP zone and a Web Unlocker zone

Setup

pnpm run setup          # install frontend deps, sync the Python agent, create env files
lk app env -w agent-py  # write LiveKit creds into the agent
lk app env -w frontend  # write LiveKit creds into the frontend

Add Moss + Bright Data credentials to agent-py/.env.local (see agent-py/.env.example for the full list):

MOSS_PROJECT_ID=your_moss_project_id
MOSS_PROJECT_KEY=your_moss_project_key
BRIGHT_DATA_API_TOKEN=your_bright_data_token
BRIGHT_DATA_SERP_ZONE=your_serp_zone
BRIGHT_DATA_UNLOCKER_ZONE=your_unlocker_zone

Run

pnpm run moss:index   # seed the Moss knowledge + memory indexes (once)
pnpm dev              # start the voice agent + frontend together

• Frontend: http://localhost:3000 → Start voice demo
• Try: "Find me the cheapest Sony WH-1000XM5" then "Watch it and tell me if it drops below 130."

Scripts

Script	Purpose
pnpm run setup	Install deps and create local env files
pnpm dev	Run the Python voice agent and Next.js frontend together
pnpm run moss:index	Seed the Moss indexes
pnpm run agent:py:console	Run the voice agent in terminal console mode
pnpm test	Run the Python tests
pnpm lint	Lint the agent and frontend

---

Built at AI Hack Day. Inspired by the LiveKit + Moss + Bright Data starter, reworked into a voice shopping use case.