Overview
Artificial intelligence is woven throughout the High IQ platform, powering features that would be impossible to deliver manually at scale. From reading dispensary labels with computer vision to generating personalized consumption reports in real time, AI handles the heavy lifting so you can focus on making informed decisions. This page covers the four primary AI-powered capabilities in High IQ and the technology behind each.Label Scanner
The AI Label Scanner is one of High IQ’s signature features. Point your phone’s camera at a dispensary label — whether it is a Certificate of Analysis (COA) sticker, retail packaging, or a jar label — and AI extracts structured data in seconds.What It Extracts
| Data Point | Description |
|---|---|
| Strain name | Identified from label text, matched against the database |
| THC / CBD percentages | Extracted from lab results or packaging claims |
| Terpene profile | Individual terpene names and concentrations (mg/g) when available |
| High Family | Classified into one of the 6 High Spectrum Families based on extracted terpenes |
| Product type | Flower, concentrate, edible, cartridge, etc. |
How It Works
Image Capture
You photograph the label using the in-app scanner. The scanner supports single images or batch mode (up to 4 images for front, back, COA, and additional angles of the same product).
AI Vision Analysis
The image is sent to Google Gemini 3 Flash, a multimodal AI model optimized for visual document understanding. The model reads printed text, tables, charts, and even handwritten notes on labels.
Data Extraction
The AI extracts structured fields — strain name, cannabinoid percentages, terpene names and values — from the unstructured label image. For batch scans, terpene profiles are merged using the maximum value found across all images.
High Family Classification
Extracted terpene data is run through the High Spectrum classification algorithm to assign the product to a High Family.
Batch Scanning
Cannabis products often split information across multiple labels. Batch mode lets you scan up to 4 images of the same product:- Front label — Strain name, brand, product type
- Back label — Cannabinoid percentages, warnings, ingredients
- COA sticker — Full terpene profile with lab-verified concentrations
Privacy
Label scanner data is handled with strict privacy controls:- Images are not stored — After AI analysis, the image is discarded. We do not retain your photographs.
- IP addresses are hashed — SHA-256 hashing ensures your IP address is anonymized before any scan metadata is stored.
- Scan data is anonymized — Extracted strain data contributes to aggregate database quality but is not linked to your personal account in any identifiable way.
The label scanner works best with clear, well-lit photographs. COA labels with printed terpene tables yield the most detailed extractions. Blurry or low-contrast images may result in incomplete data.
Streaming Reports
High IQ generates personalized consumption reports using AI, delivered in real time through server-sent event (SSE) streaming. Rather than waiting for a complete report to generate, you see each section appear as it is written — similar to watching someone type.Report Capabilities
Reports analyze your personal data from the High IQ app to provide insights on:- Consumption patterns — Frequency, timing, preferred strains, and trends over time
- Spending analysis — Total spend, cost per session, dispensary comparisons, budget optimization suggestions
- Strain preferences — Which terpene profiles, effects, and High Families you gravitate toward
- Recommendations — AI-suggested strains based on your history, preferences, and unexplored categories
Streaming Architecture
Reports use AI SDK 6 with server-sent events for real-time delivery:- Section-by-section generation — Each report is divided into logical sections (overview, consumption, spending, recommendations). The AI generates one section at a time.
- Progressive rendering — As each token is generated, it streams to your device immediately. You see text appear word by word.
- Progress tracking — A progress bar shows overall report completion with per-section status indicators.
- Resumability — If your connection drops mid-report, reconnecting picks up where generation left off.
Deep Research
Professor High, the in-app AI assistant, uses autonomous tool-calling to conduct deep research on cannabis topics. Unlike simple question-and-answer chatbots, Professor High can:- Search the strain database — Look up specific strains, compare terpene profiles, find similar strains
- Cross-reference research — Pull from the research paper database to cite published studies
- Chain multiple queries — Answer complex questions that require gathering data from several sources before synthesizing a response
Example Questions
| Question | What Happens Behind the Scenes |
|---|---|
| ”What strains are high in caryophyllene and good for pain?” | Searches database by terpene + effect, ranks results, returns top matches with explanations |
| ”How does myrcene affect sleep?” | Queries research papers, synthesizes findings, cites sources |
| ”Compare Blue Dream and Green Crack” | Fetches both profiles, generates side-by-side terpene and effect comparison |
| ”What should I try if I like OG Kush?” | Analyzes OG Kush’s profile, runs similarity search, filters by availability |
How It Differs from Generic AI
Professor High is not a generic language model answering cannabis questions from training data. It has:- Live database access — Answers are based on current strain data, not stale training knowledge
- Tool autonomy — The AI decides which tools to call and in what order, adapting its research strategy to each question
- Cannabis domain expertise — Prompts are tuned with cannabis-specific knowledge from the
@tiwih/cannabisand@tiwih/ai-promptspackages - Source citations — When referencing research, the AI links to specific papers in the research hub
AI-Generated Content
High IQ uses AI to generate educational content at scale:Blog Articles
The platform’s blog features AI-generated cannabis education articles covering strain reviews, terpene deep dives, consumption guides, and industry news. Each article goes through:- AI writing — Long-form content generated with cannabis domain context
- Image generation — Thumbnail images created by Google Gemini with cannabis-appropriate visual styles
- Human review — All AI-generated content is reviewed for accuracy before publication
Strain Descriptions
When a new strain enters the database through the research pipeline, AI generates:- Comprehensive descriptions — Multi-paragraph overviews covering genetics, effects, and recommended use cases
- Effect summaries — Concise effect profiles for quick scanning
- Growing notes — Cultivation information synthesized from multiple sources
Research Paper Summaries
The daily paper pipeline uses Claude to generate plain-language summaries of published cannabis research. These summaries make academic papers accessible to consumers who want science-backed knowledge without reading full journal articles.AI Models Used
| Feature | Model | Provider | Why This Model |
|---|---|---|---|
| Label Scanner | Gemini 3 Flash | Google (via AI Gateway) | Fast multimodal vision for document understanding |
| Streaming Reports | Claude Opus 4.6 (creative), Claude Sonnet 4.5 (analysis), GPT-5-mini (structured) | Multiple (via AI Gateway) | Tiered model selection by section complexity |
| Deep Research (Professor High) | Claude Sonnet 4.5 | Anthropic (via AI Gateway) | Best for agentic research with autonomous tool calling |
| Paper Summaries | Claude Sonnet 4.5 | Anthropic (via AI Gateway) | Native structured outputs for reliable JSON generation |
| Paper Quality Gate | Claude Sonnet 4.5 | Anthropic (via AI Gateway) | Native structured outputs for quality assessment |
| Blog Writing | Claude Opus 4.6 | Anthropic (via AI Gateway) | Best for creative long-form content |
| Blog Images | Gemini 3 Pro | High-quality native image generation | |
| Receipt Parsing | Mistral Small | Mistral (via AI Gateway) | Native JSON Schema for accurate structured extraction |
| Strain Matching | GPT-5-mini | OpenAI (via AI Gateway) | Fast, cost-effective matching logic |
apps/api/src/config/ai.ts) so swapping a model for any feature requires changing only one line.
Responsible AI Practices
High IQ follows these principles in its AI implementation:- Transparency — AI-generated content is labeled as such. Users know when they are reading AI output.
- Accuracy over speed — Quality gates and human review catch AI hallucinations before content reaches users.
- Privacy first — User data processed by AI for reports is not used to train models or shared with third parties.
- No medical claims — AI is instructed to never make specific medical claims or recommend cannabis as treatment for conditions. It provides educational information only.
- Continuous improvement — We monitor AI output quality and update prompts and models as better options become available.
Frequently Asked Questions
Is my data used to train AI models?
Is my data used to train AI models?
No. Your personal consumption data, stash information, and label scans are used only to generate your personal reports and contribute anonymized aggregate data to the strain database. Your data is never sent to AI providers for model training.
How accurate is the label scanner?
How accurate is the label scanner?
Accuracy depends heavily on image quality and label clarity. COA labels with printed terpene tables achieve the highest extraction rates. Retail packaging with large text and clear layouts also performs well. Handwritten or low-contrast labels may require manual correction.
Can I use Professor High for medical advice?
Can I use Professor High for medical advice?
No. Professor High is an educational tool. It can help you understand terpene science, compare strain profiles, and find research papers, but it cannot and should not replace professional medical advice. Always consult a healthcare provider for medical decisions.
Why do reports stream instead of loading all at once?
Why do reports stream instead of loading all at once?
Streaming provides a better user experience. AI report generation takes 30-60 seconds for a full report. Streaming lets you start reading immediately rather than staring at a loading spinner. It also allows the app to display progress so you know exactly where generation stands.