Back to corpus
architecturetechnical paper candidatescore 74

BWB Kiosk — Voice Ordering Architecture

> *"Break every component down to its grills... define a subsection and a sub-subsection that further builds upon the previous section, then expands it in a recursive manner."*

Open full HTMLDownload HTMLCorpus JSONPaper schema

Full HTML reader

Read the full artifact

Open in new tab

Extracted abstract or opening context

# BWB Kiosk — Voice Ordering Architecture ### Deep Recursive Decomposition & Evolutionary Design *v2.0 — February 10, 2026 — Verified against codebase* > *"Break every component down to its grills... define a subsection and a sub-subsection that further builds upon the previous section, then expands it in a recursive manner."* 1. [Vision & Philosophy](#1-vision--philosophy) 2. [System Topology](#2-system-topology) 3. [Layer 1: Audio Foundation](#3-layer-1-audio-foundation) 4. [Layer 2: Speech-to-Text Pipeline](#4-layer-2-speech-to-text-pipeline) 5. [Layer 3: Natural Language Understanding](#5-layer-3-natural-language-understanding) 6. [Layer 4: Dialogue Engine](#6-layer-4-dialogue-engine) 7. [Layer 5: Order State Machine](#7-layer-5-order-state-machine) 8. [Layer 6: Synthesis & Feedback](#8-layer-6-synthesis--feedback) 9. [Layer 7: Interaction Surface](#9-layer-7-interaction-surface) 10. [Layer 8: Learning & Telemetry](#10-layer-8-learning--telemetry) 11. [Cross-Cutting: Error Taxonomy & Recovery](#11-cross-cutting-error-taxonomy--recovery) 12. [Cross-Cutting: Performance & Latency Budget](#12-cross-cutting-performance--latency-budget) 13. [Cross-Cutting: Privacy, Security, Offline](#13-cross-cutting-privacy-security-offline) 14. [State Machine Formal Specification](#14-state-machine-formal-specification) 15. [End-to-End Flow Traces](#15-end-to-end-flow-traces) 16. [Gap Analysis: Current vs Next-Gen](#16-gap-analysis-current-vs-next-gen) 17. [Evolution Roadmap](#17-evolution-roadmap) 18. [Codebase Map (Verified)](#18-codebase-map-verified) ### 1.1 Core Vision A voice ordering system that feels like the best barista conversation — one that understands messy human speech, self-corrections, contextual references, and emotional undertones — delivered at kiosk speed with zero training required. #### 1.2.1 Conversational Intelligence ##### [ip] Natural Speech Tolerance - **Disfluency handling**: "um", "uh", "like", "so" stripped by `TranscriptNormalizer` (228 LOC) - **Self-correction**: "large, no wait, medium" — correction markers detected: "actually", "scratch that", "wait no", "I mean", "not that" - **Partial utterances**: Streaming partial transcripts handled by `TranscriptPipeline` (188 LOC) — deduplicate "med" → "medi" → "medium" - **Run-on orders**: "a latte and two cappuccinos and oh also a muffin" — multi-item extraction via `EntityExtractor`

Promotion decision

What has to happen next

Promote into a technical note or architecture paper with implementation anchors.

Why this is not always a full paper yet

Corpus pages are public-safe readers for discovered workspace artifacts. They are not automatically final papers. A corpus item becomes a polished paper only after the editable source, evidence checkpoints, references, figures, render path, and release status are attached through the paper schema.