Pune Plate Agent

Inspiration

Pune's food scene is spread across neighbourhoods, and generic ratings do not capture personal taste. Pune Plate Agent plans a food crawl around any Pune neighbourhood from restaurants you already love.

What it does

You name an area, a few favourite restaurants, and optionally a diet and a craving. It asks Qloo's taste graph for venues with affinity to your anchors, filters out pure lodging, ranks by taste affinity (not distance), and lays out a crawl with reasons and tag evidence. A rule-based planner covers the standard flow; an AI agent mode answers free-form requests.

How we built it

A Cloudflare Worker (ESM, no dependencies) serves the web UI and a JSON API. Qloo's Insights API supplies taste-matched venues within about 2.5 km of the chosen area (widened to 6 km when matches are sparse, and the UI says so). The AI agent drives an OpenAI-compatible LLM with tool calls under an explicit call/neuron budget.

Challenges and what we learned

Taste affinity is not geographic containment: top picks can sit outside the named neighbourhood, so the app states the actual search radius rather than implying otherwise. Craving keywords are enforced in the rule planner; the AI mode does not enforce them yet, and the app discloses that limitation.

Built With

• cloudflare-workers
• javascript
• llm
• qloo