ElixirConf 2023 Embedded System with Elixir for millions AI-based point of sales

LiveView was the entry point as we needed a real-time dashboard monitoring our point of sales. We extended all the Core through Phoenix, including WebSockets for IoT devices. Finally, we developed our firmware with Nerves, and we are exploring evision for AI inference. We want to share our history and show some examples of security, simplicity, and improvements in our tech based on the Elixir ecosystem.

Transcript

1. Building Embedded and Core Systems with Elixir for millions AI-based points of purchase Alfonso • 10+ yrs working with software • 6 yrs in the Elixir ecosystem • 3 years working to build our platform • phonnz
3. ➔ Liveview Dashboard ➔ Remote firmware updates ➔ Nerves ➔ Remote control ➔ Deployment ➔ Clostering ➔ Monitoring ➔ Dream of the product ➔ The business ➔ Tech challenges ➔ First architecture Agenda ➔ AWS deployments ➔ Custom System ➔ Custom Wizards ➔ AWS IoT Core Greengrass ➔ 5G network How it started? How Elixir helped us? Learnings & Tips
4. Running in a low cost device Commercial Wallet integration Mobile availability On-site data collection Cloud Data sending Make a purchase Inference Charge the customer Register user & payment-method On-site operation Have a way to infer the products Charge Business requirements Technical requirements
6. Frist System Arch Diagram
7. Websockets Client
8. Linux service
9. Git
11. ➔ Liveview Dashboard ➔ Remote firmware updates ➔ Nerves ➔ Remote control ➔ Deployment ➔ Clostering ➔ Monitoring Agenda How Elixir helped us?
12. ➔ Limited resources ➔ Slow Startup ➔ Slow communication ➔ Corrupted updates ➔ Limited USB bandwidth ➔ Power supply ➔ Current consumption Performance Issues
13. Phoenix Liveview Dashboard
14. PM2 Concurrency
15. ➔ 3 video capturing ➔ 2 devices ➔ Communication between devices ➔ Video Sync ➔ *learning
16. Rusty Version
17. Technical requirements (still) ★ Running in a low cost device ★ Remote everything
22. ➔ gstreamer ? ➔ Omx encoder ? ➔ Nerves back with 32bits ? ➔ Membrane ? Nice! But we need video recording!
23. We tried… ➔ Mix release & PIOS ➔ Nerves + Docker ➔ Nix
24. We tried… ➔ Mix release & PIOS ➔ Nerves + Docker ➔ Nix
25. We liked…
28. Agenda ➔ AWS deployments ➔ Custom System ➔ Custom Wizards ➔ AWS IoT Core Greengrass ➔ 5G network Learnings & tips
29. ➔ More FPS ➔ High resolutions ➔ Faster startup ➔ Encode Still need to improve video recording!
31. Adding packages. Using search *rtl8188eu wifi adapter
33. When updating Kernel: check diff and build
34. Brand new box Actual Firmware Recording high quality & high FPS Processing first step inside ARM Custom ARM-based board
35. Greengrass
36. thing = Device thing group = group of things artifact = Installable bundle core device = Greengrass id Deployment = Workflows to make possible the upgrade Greengrass
37. Requirements: - Valid Identity - Check for Artifacts - Download/Install - Run - Keep it safe Greengrass
41. Building firmware in the cloud
42. Component: Logic group with recipe and configs Recipe: metadata, configuration parameters,dependencies, lifecycle, and platform compatibility. Artifact: binary bundle (scripts, compiled code, statics, files). Component Recipe
43. Component Recipe
45. Allocate Greengrass daemon
46. Allocate Greengrass daemon
47. Run Greengrass daemon
48. Run Greengrass daemon
49. Run Greengrass daemon
50. Greengrass daemon Diagram explaining Greengrass daemon workflow - Has certificates - Provide new certificates - Look for new artifacts (pending deployment) - Download artifact (bundle) - Run fwup
51. 5G
52. Multiregion
53. Multiregion
54. We did it !
55. We did it ! ➔ Support Million Devices ➔ Real-Time status ➔ Remote everything ➔ Multiplatform Firmware ➔ Compatibility ➔ Enabled teams to experiment
56. We did it ! ➔ Support M Devices ➔ Real-Time status ➔ Remote ➔ Multiplatform Firmware ➔ Team
57. What we learn Train your Team Hardware is not easy, Nerves is really accessible though Choose your tools, and trust the expertise Get & Give support from community Replace / Migrate strategies Have hardware experts Study stack capabilities Measure Collect Data
58. Q&A Alfonso Gonzalez alfonso@intuitivo.com phonnz • intuitivo.com intuitivo.com/careers Linkedin.com: intuitivo-ai •