Shopping Agent

​

Overview

​

What It Does

• Natural language search — Ask in plain English: “Find Gorilla Glue near me,” “Cheapest eighth of Wedding Cake within 15 miles,” “Something good for focus and creativity under $50”
• Multi-dispensary menu search — Checks menus at all dispensaries within your specified radius simultaneously
• Price comparison — Shows the same strain at different dispensaries with pricing so you can compare
• Availability confidence — Indicates how fresh the availability data is: “Confirmed in stock (updated 2 hours ago)” vs. “Likely in stock (menu data from 3 days ago)”
• Alternative suggestions — If the exact strain is not available, suggests similar strains that are: “OG Kush is not in stock nearby, but these 3 strains share 80%+ terpene similarity and are available at dispensaries within 5 miles”
• Budget-aware — Specify a budget and the agent filters and sorts by price-to-quality ratio
• Effect-based shopping — “I need something for sleep” is translated into terpene and High Family preferences and matched against available inventory
• Professor High personality — The agent responds in Professor High’s voice: knowledgeable, enthusiastic, and opinionated. “I found Wedding Cake at Green Thumb for $45/eighth — solid price for a strain with this caryophyllene load.”
• Directions — Tap to navigate to the recommended dispensary with one tap
• Save to wishlist — Add any suggested strain to your wishlist for future reference
• Purchase follow-up — After you visit, the agent prompts: “Did you find what you were looking for? Log it to your stash.”

​

User Value

​

How It Works

​

Technical Approach

​

Agent Architecture

User Query → Professor High (AI SDK 6 with tool calling)
                      ↓
              Intent Classification
              (strain search, effect search, budget search, restock)
                      ↓
              Tool Calls (parallel where possible):
              ├── Strain Database Search (existing API)
              ├── Dispensary Menu Query (new API)
              ├── Price Comparison Engine (new)
              ├── User Profile Lookup (existing Convex)
              └── Geolocation (existing)
                      ↓
              Result Scoring & Ranking
                      ↓
              Response Generation (Professor High voice)
                      ↓
              Structured Result Cards + Conversational Text

​

Tool Definitions

​

Dispensary Menu Data

​

Conversation Memory

• Previous queries — “Show me something cheaper” refers to the current results
• User preferences — Loaded once at session start from the user’s profile
• Dispensary context — If the user asks about a specific dispensary, subsequent queries are scoped to that location
• Session state — The agent remembers which results the user has already seen

​

Fallback Strategies

​

Background Processing

• Streaming response — Professor High narrates the search process: “Checking menus at 12 dispensaries near you… Found OG Kush at 3 locations… Comparing prices…”
• Progressive results — Results appear as they are found, not all at once
• Background task option — For very broad searches, the agent can run in the background and send a push notification when results are ready

​

Potential Trigger.dev Integration

​

Tier Impact

​

Dependencies

• Professor High AI chat infrastructure — built and live
• AI SDK 6 with tool calling — built and live
• Strain database and similarity search — built and live
• Dispensary directory in Convex — built and live
• Trigger.dev for background tasks — built and live
• Dispensary menu data partnerships (Dutchie, Jane, or equivalent)
• Menu data caching and freshness tracking infrastructure
• Price comparison and normalization engine
• Shopping-specific tool definitions for the AI agent
• Result card UI component
• Conversation memory for multi-turn sessions
• Dispensary Map integration (planned)
• Smart Suggestions integration for alternative recommendations (planned)
• Push notification system for proactive alerts (optional)

​

Open Questions

1. Menu data partnerships — This is the single biggest dependency and risk. Without reliable dispensary menu data, the agent cannot check availability or compare prices. Dutchie and Jane are the largest platforms, but partnerships require business development effort. Should we launch with user-reported availability (crowd-sourced) as an MVP while pursuing partnerships?
2. Proactive vs. reactive — Should the agent be reactive only (you ask, it searches), or also proactive (it monitors and alerts you when something becomes available)? Proactive is much more valuable but requires background task infrastructure and push notifications.
3. Voice interface — Should the Shopping Agent support voice input? “Hey Professor High, find me some OG Kush” is a natural interaction for someone standing in a dispensary. Voice adds significant complexity (speech-to-text, noise handling) but could be a differentiator.
4. Affiliate revenue — Could the Shopping Agent generate revenue through dispensary referral fees? “We sent 50 customers to your dispensary this month.” This is a potential business model but must not compromise recommendation objectivity.
5. Legal considerations — Facilitating cannabis purchases involves legal complexity. Are we “recommending” a purchase or “providing information”? Legal review needed on the agent’s language and disclaimers, especially across state lines.

​

Related Features

• Dispensary Map — The spatial foundation that the Shopping Agent queries
• Smart Suggestions — The recommendation engine that powers alternative suggestions
• Professor High — The AI personality that the Shopping Agent embodies
• Dispensary Tracking — User’s saved dispensaries are prioritized in agent results
• Session Journal — Post-purchase logging closes the feedback loop