Upload community_contributions/damola-adewunmi/README.md with huggingface_hub

community_contributions/damola-adewunmi/README.md

@@ -0,0 +1,51 @@
+# Deal Hunter — PlayStation UK Gift Cards
+
+Multi-agent **Deal Hunter** for PlayStation UK gift card prices, built with **LangGraph**. No local fine-tuned model; **GPT-4o** is used for all reasoning, with a specialist system prompt in the Valuer to simulate pricing expertise.
+
+## Workflow
+
+1. **Searcher Agent** — Finds current UK gift card prices via **Tavily** (CDKeys, ShopTo, Eneba, GAME, etc.).
+2. **Valuer Agent** — **GPT-4o** with a specialist system prompt:
+   - PlayStation UK cards usually retail at **5–10%** discount.
+   - **Strong Buy** = discount **>12%**.
+3. **Strategist Agent** — Summarizes findings into a **Gradio** table.
+
+## Setup
+
+From the **repo root** (so `langgraph`, `langchain-openai`, etc. are available):
+
+```bash
+# Optional: add Tavily for live search (otherwise placeholder results)
+export TAVILY_API_KEY=your_key
+export OPENAI_API_KEY=your_key
+
+cd week8/community_contributions/damola-adewunmi
+uv run app.py
+```
+
+Or install deps in this folder and run:
+
+```bash
+pip install -r requirements.txt
+python app.py
+```
+
+## Project layout
+
+- `state.py` — LangGraph shared state (`DealHunterState`).
+- `graph.py` — Builds the graph: **searcher → valuer → strategist**.
+- `prompts.py` — **Valuer specialist system prompt** (expertise rules, JSON output).
+- `agents/searcher.py` — Tavily search node.
+- `agents/valuer.py` — GPT-4o + specialist prompt; parses JSON deals.
+- `agents/strategist.py` — Builds table rows for the UI.
+- `app.py` — Gradio UI and pipeline entrypoint.
+
+## Valuer prompt (expertise)
+
+The Valuer uses a detailed system prompt in `prompts.py` that:
+
+- Defines **Strong Buy** (>12% discount), **Good/Fair** (5–10%), **Weak/Avoid** (<5% or above face).
+- Requires **JSON** output with `retailer`, `denomination_gbp`, `price_gbp`, `discount_pct`, `verdict`, `url`.
+- Instructs the model to use **only** information from the provided search results (no hallucination).
+
+This mimics the behaviour of a fine-tuned pricing specialist without training a custom model.