How It Works
How Pre-Shift Works
Restaurant truth becomes adaptive training through a 7-layer pipeline that improves with every staff interaction.
KNOWLEDGE INGESTION
PROCESSING PIPELINE
CANONICAL KNOWLEDGE BASE
GENERATION ENGINE
TRAINING MODALITIES
MEASUREMENT & MASTERY
CLOSED-LOOP IMPROVEMENT
KNOWLEDGE INGESTION
Menu Upload
Food · Cocktails · Wine
Training Docs
Core · Role · Reference
URL Monitoring
Auto-detect changes
Raw content
PROCESSING PIPELINE
Vision Extraction
GPT-4 per-page OCR
Enrichment
Descriptions · Pairings
Topic Discovery
Auto-tag from content
Vectorization
Pinecone embeddings
Structured knowledge
CANONICAL KNOWLEDGE BASE
Food KB
Items + allergens
Cocktail KB
Builds + specs
Wine KB
BTG + Bottles
Dossier
Docs + policies
Grounded context
GENERATION ENGINE
Golden Path
Single-item grounding
Rubric Builder
Weighted key points
Similar Items
Vector context map
Self-Critique
Quality gate
Questions + rubrics
TRAINING MODALITIES
Modules
Structured curriculum
Assessments
Measure knowledge
Diagnostic
Adaptive CAT
Build a Lesson
On-demand learning
Challenge
Weakness targeting
Ask Pre-Shift
RAG Q&A assistant
Attempts + answers
MEASUREMENT & MASTERY
Semantic Grading
Deterministic + LLM
ELO Ratings
Staff + question difficulty
Topic Mastery
Granular per-topic
Staff Feedback
Thumbs up/down/flag
Performance signals
CLOSED-LOOP IMPROVEMENT
Calibration Agent
8 weekly analyses
Generation Brief
Exemplars + gaps
Quality Audit
Flag + retire weak Qs
Misconception Mining
From real answers
CLOSED LOOP — Performance signals refine generation continuously
7
Pipeline layers
6
Training modalities
8
Calibration analyses
100%
Question grounding
See the Full Architecture
Every connection is real. Every path is built. Explore how data flows between each component of the system.
INPUTS
PROCESSING
KNOWLEDGE
GENERATION
TRAINING
MEASUREMENT
CLOSED LOOP
INPUTS
Menu PDF
or URL
Training Doc
Role · Core
Live URL
Auto-monitored
PROCESSING
Vision Extract
GPT-4 per page
Sync Engine
Diff · Archive · Adopt
KNOWLEDGE
Knowledge Base
Food · Cocktail · Wine · Dossier
Enrichment
Notes · Pairings
Vectors
Pinecone
Topics
Auto-discovered
GENERATION
Golden Path
Grounded generation
Rubric Builder
Weighted key points
Similar Items
Cross-menu context
TRAINING
Modules
Structured
Assessments
Evaluate
Diagnostic
Adaptive
Lessons
On-demand
Challenge
Weakness
Ask PS
RAG Q&A
MEASUREMENT
Semantic Grading
Partial credit + LLM arbiter
ELO + Mastery
Per-topic · Per-staff
Staff Signals
Thumbs up/down/flag
CLOSED LOOP
Calibration Agent
8 weekly analyses
Generation Brief
Exemplars · Gaps
CLOSED LOOP — Performance signals refine generation continuously
Not a quiz maker. A training compiler.
Every question traces to a real menu item. Every answer improves the next cycle. The system doesn't just train — it learns how to train better.