Build a voice based Kiosk for Retail store using Moorcheh-edge and Arduino Uno Q

Arduino App Lab

3.5 mm Lavalier Microphone

3.5 mm Screw Terminal Adapter

UGREEN USB Sound Card (Mic + Headphone/Line Out)

Moorcheh-edge

UGREEN Revodok Pro USB-C Hub 6-in-1 (100W PD)

Arduino® UNO™ Q 4GB

<html><body><h3>Retail Kiosk with Moorcheh Edge - On-Device AI Assistant for Arduino UNO Q</h3><p>Turn Arduino UNO Q into a privacy-first retail assistant: local vector search, on-device LLM answers, and USB voice - no cloud API keys. Provision once, then run from USB-C power with your store knowledge on the board.</p><h3>Overview</h3><p>This project builds a retail in-store assistant on the Arduino UNO Q using Moorcheh Edge - an edge-AI stack for semantic search, retrieval-augmented generation (RAG), and hands-free voice on constrained ARM hardware.</p><p>Shoppers ask natural-language questions about products, prices, hours, and policies. The UNO Q searches a local knowledge base, generates answers with a small on-device LLM, and (with a USB mic and speaker) listens and speaks - without sending customer questions to the cloud.</p><p>The reference demo uses a neighborhood café and mini-market (The Brew Corner), but the same pattern works for grocery, pharmacy, boutique, or any fixed-location retail format: you change the uploaded content, not the platform.</p><p>Powered by Moorcheh - PyPI package moorcheh-edge, Docker image moorcheh/moorcheh-edge, and documentation at <a href="https://docs.moorcheh.ai/on-edge/examples/retail-kiosk">docs.moorcheh.ai/on-edge</a>.</p><h3>Why Moorcheh</h3><p>Moorcheh is the world's first information-theoretic search engine. Instead of storing full-precision vectors, it compresses embeddings from FP32 down to 1-bit representations. That dramatically cuts storage and RAM while maintaining the same search accuracy you expect from semantic retrieval.</p><p>On edge hardware this matters:</p><ol><li>~17 ms search over 10,000 vectors - fast enough for real-time kiosk Q&amp;A</li><li>Significantly lower RAM than conventional vector databases - room left on UNO Q for Ollama, voice, and embeddings</li><li>Purpose-built for on-device deployment - why Moorcheh Edge runs comfortably on ARM boards like Arduino UNO Q where generic cloud-first vector stacks do not</li></ol><p>Retail kiosks need search + RAG + voice in a few gigabytes of RAM. Moorcheh's compression and speed are what make that stack practical at the edge.</p><h3>Why edge AI for retail?</h3><ol><li>Privacy - customer questions and store data stay on the device</li><li>Reliability - no dependency on internet uptime for answering</li><li>Cost - no per-query cloud LLM or embedding API fees</li><li>Latency - search and generation run on local ARM64 Linux</li><li>Trust - answers grounded in your uploaded catalog (RAG), not open-ended hallucination</li></ol><p>What runs on Arduino UNO Q</p><p>All inference and voice workloads run on the UNO Q Linux (MPU) side:</p><ol><li>Moorcheh Edge (Docker, port 8080) - vector store, similarity search, streaming RAG API</li><li>Ollama (port 11434) - local LLM <code class="code">llama3.2:1b-instruct-q4_K_M</code> (quantized, ~808 MB)</li><li>moorcheh-edge voice serve (port 8766) - speech-to-text (Whisper), text-to-speech (Piper), mic/speaker bridge</li><li>BGE embeddings - client-side text embeddings (768-dim) for catalog and queries</li><li>Persistent store - <code class="code">~/.moorcheh-edge/data</code>, survives reboots</li></ol><p>A display PC on the same Wi-Fi can run a React touchscreen UI (retail-kiosk-api + browser). The AI brain stays on the UNO Q; the PC is optional for the full kiosk screen experience.</p><h3>Standalone operation after provisioning</h3><p>Once you complete one-time setup and load your store catalog, the UNO Q can act as a self-contained edge appliance.</p><p><strong>One-time setup (requires a PC and network access):</strong></p><ol><li>Install <code class="code">moorcheh-edge</code> in a Python venv on the UNO Q.</li><li>Run <code class="code">moorcheh-edge up --with-llm --warm-llm</code> to start Docker, Ollama, and the answer model.</li><li>Run <code class="code">moorcheh-edge voice setup</code> (USB mic and speaker).</li><li>Upload your store catalog (demo script or Admin sync from the PC).</li><li>Start <code class="code">moorcheh-edge voice serve</code> for HTTP voice endpoints.</li></ol><p><strong>Day-to-day operation:</strong></p><ol><li>Power the board with a USB-C charger (standard phone/tablet adapter or powered USB port).</li><li>No cloud API keys and no ongoing internet requirement for customer Q&amp;A - knowledge and models live on the device.</li><li>Wi-Fi is only needed when you update catalog content from a staff PC or run the touchscreen UI.</li><li>After power-on, activate the venv and start services (or configure auto-start) - then the board answers from local data.</li></ol><p>Without a display PC:</p><p><code class="code">moorcheh-edge voice ask --kiosk-mode</code></p><p>Speak into the USB mic; hear the answer on the USB speaker - entirely on the UNO Q.</p><h3>How it works</h3><figure><img src="https://projects.arduinocontent.cc/dcd667fb-32b0-438c-910f-3b2609947ca1.png"/><figcaption>Retail kiosk architecture</figcaption></figure><p></p><p>Display PC (optional) runs the browser UI (:5173) and retail-kiosk-api (:8765). Arduino UNO Q runs Moorcheh Edge (:8080), Ollama (:11434), and moorcheh-edge voice serve (:8766) with USB mic and speaker.</p><ol><li>Text questions: PC UI → API → streaming RAG on UNO Q (/answer/stream).</li><li>Voice questions: PC UI → voice proxy on UNO Q → RAG + TTS on device hardware.</li><li>Catalog: embedded and stored on the UNO Q; updates sync over Wi-Fi when staff edit content.</li></ol><p>Hardware wiring: <a href="https://www.figma.com/board/YiBj2eVpuqvkqL2RNyDmp2/Retail-Kiosk-with-Moorcheh?node-id=0-1&amp;p=f&amp;t=RBODYlsZiaJjI3Sj-0">Figma guide</a>.</p><h3>Features</h3><p><strong>Customers</strong></p><ol><li>Multi-turn conversational Q&amp;A grounded in store documents</li><li>Streaming text answers on a touchscreen (with display PC)</li><li>Push-to-talk voice via UNO Q mic and speaker</li><li>Kiosk-style relevance filtering (kiosk_mode) for in-aisle use</li></ol><p><strong>Store staff</strong></p><ol><li>Admin UI to add, edit, and delete product and policy documents</li><li>Editable LLM system and footer prompts</li><li>Chunk-level view of what is synced to the edge store</li><li>Demo catalog (The Brew Corner) included for evaluation</li></ol><p>Developers</p><ol><li>Open REST APIs: Moorcheh Edge (:8080) and voice server (:8766)</li><li>Python CLI: moorcheh-edge up, search, answer, voice serve</li><li>Multi-arch Docker image (linux/arm64 for UNO Q, linux/amd64 for PC dev)</li><li>Documentation and quick start in the GitHub repository</li></ol><h3>Hardware requirements</h3><ol><li>Arduino UNO Q - ARM64 Linux (~3.6 GB RAM), Docker capable</li><li>USB-C power - standard 5 V supply; board runs as edge appliance after setup</li><li>USB microphone - voice input (ALSA-compatible)</li><li>USB speaker - voice output</li><li>Wi-Fi - required for initial provisioning and optional PC UI; not required for offline Q&amp;A after catalog load</li><li>Display PC (optional) - Windows, Linux, or macOS, for touchscreen kiosk only</li></ol><h3>Software stack</h3><ol><li>Edge vector DB + RAG API - Moorcheh Edge (Rust, Docker)</li><li>Embeddings - BGE-base-en-v1.5 (768-dim, client-side)</li><li>LLM - Ollama + Llama 3.2 1B Instruct Q4</li><li>STT - Whisper (whisper.cpp, tiny.en)</li><li>TTS - Piper (with espeak-ng fallback)</li><li>Kiosk API + Admin - Python FastAPI (retail-kiosk)</li><li>Customer UI - React + Vite</li></ol><h3>Getting started</h3><ol><li>Clone github.com/moorcheh-ai/retail-kiosk</li><li>Follow QUICKSTART.md for UNO Q provisioning and PC UI setup</li><li>Read docs.moorcheh.ai/on-edge</li><li>Arduino guide: docs.moorcheh.ai/on-edge/guides/arduino-uno-q</li></ol><p>Minimum UNO Q commands after install:</p><p></p><p><code class="code">source ~/moorcheh-venv/bin/activate</code></p><p><code class="code">moorcheh-edge up --with-llm --warm-llm -y</code></p><p><code class="code">moorcheh-edge voice setup</code></p><p><code class="code">python ~/upload-catalog-to-edge.py --catalog ~/brew-corner-catalog.json -y</code></p><p><code class="code">moorcheh-edge voice serve --port 8766</code></p><h3>Use cases</h3><ol><li>In-aisle product and policy questions (gluten-free options, return policy, hours)</li><li>Hands-free voice kiosk at a café counter or mini-market</li><li>Privacy-sensitive retail without cloud inference</li><li>Edge AI proof-of-concept on Arduino UNO Q before custom enclosure deployment</li><li>Fixed-location assistants where content changes but hardware stays the same</li></ol><h3>Links</h3><ol><li>Retail kiosk - https://github.com/moorcheh-ai/retail-kiosk</li><li>Moorcheh Edge - https://github.com/moorcheh-ai/moorcheh-edge</li><li>Documentation - https://docs.moorcheh.ai/on-edge/</li><li>PyPI - https://pypi.org/project/moorcheh-edge/</li><li>Docker Hub - https://hub.docker.com/r/moorcheh/moorcheh-edge</li><li>Hardware wiring - https://www.figma.com/board/YiBj2eVpuqvkqL2RNyDmp2/Retail-Kiosk-with-Moorcheh</li><li>Moorcheh - https://moorcheh.ai</li></ol><p></p></body></html>

Docs

Retail-Kiosk GitHub

Build a voice based Kiosk for Retail store using Moorcheh-edge and Arduino Uno Q

On-device retail kiosk on Arduino UNO Q: Moorcheh vector search, local RAG + LLM, Whisper STT, Piper TTS - private, fast, no cloud API keys. Power with USB-C after you load your catalog.