Tag: hardware

Machine Learning

Auris I – a hackable ultra low-power AI pin

This short article is just to share a new side project I started working on during the 2025/2026 break. Over the past year, we have started to see a vast number of companies developing hardware for AI assistants. There were also a lot of acquisitions related to these devices (e.g., Amazon bought Bee [1]), and there were also a lot of fiascos, of course (e.g., Rabbit devices, Humane, and so on). I think that these devices have a strong future ahead due to their potential, so I started prototyping some code for recording, compression, and transmission, and testing some microphones. I found excellent transcription results even with a single microphone, so I decided to start a small side project which ended up becoming the the Auris I on the right.

Routing took quite a bit of work, as the main SoC used not only castellated pins but also pins under the chip which were very difficult to route. Another issue was the requirements of the RF module, which required some clearance on the ground plane, however, this was still much easier than going straight with the RF SoC without the guest board. The main challenges now are chip shortages for some components, but I expect to have a prototype working by the end of February. I will do another post then with more details and some interesting results. The board ended up being only 5 cm. There is, of course, a lot that can be reduced, but that will come in the Auris II after I manage to hack it a bit.

Update 7/Feb: Manufacturing, PCBs arrived !

Auris I – PCB layers

I decided to go with JLCPCB with 4 layers as it is often quick and also very cheap, the only issue I faced was with the components, my board is based on nRF5340 (from Nordic Semiconductor) and modules with it were all without stock, so I had to acquire and wait for them to ship to JLCPCB, which took around 2 weeks. After that it was another week for the PCB fabrication and the assembly as well. Since the MEMS microphone was so small and the nRF5340 module also had pads under it, they required an x-ray inspection to make sure it was all connected fine. But that was around 4 pounds for x-ray and it was very fast as well. After around 5 days in total I finally received the PCBs and the quality is just amazing, I didn’t even order the cleaning of the board but they came all very clean well soldered as well.

I did all tests for the external flash, microphone, leds, battery/usb supply and also the SWD connection and it is all working fine, so now it is the fun part of coding, soon I will open-source everything, so stay tuned !

– Christian S. Perone

by Christian S. Perone
Machine Learning

TorchStation Prototype V1 – GPUs panel

I finally had some time over the holidays to complete the first panel of the TorchStation. The core idea is to have a monitor box that sits on your desk and tracks distributed model training. The panel shown below is a prototype for displaying GPU usage and memory. I’ll continue to post updates as I add more components. The main challenge with this board was power: the LED bars alone drew around 1.2A (when all full brightness and all lit up), so I had to use an external power supply and do a common ground with the MCU, for the panel I used a PLA Matte and 3mm. Wiring was the worst, this panel alone required at least 32 wires, but the panel will hide it quite well. I’m planning to support up to 8 GPUs per node, which aligns with the typical maximum found in many cloud instances. Here is the video, which was quite tricky to capture because of the camera metering of exposure that kept changing due to the LEDs (the video doesn’t do justice to how cool these LEDs are, they are very bright and clear even in daylight):

I’m using for the interface the Arduino Mega (which uses the ATmega2560) and Grove ports to make it easier to connect all of this, but I had to remove all VCCs from the ports to be able to supply from an external power supply, in the end it looks like this below:

                 ┌────────── PC USB (5V, ≤500 mA)
                 │
            +5 V │
                 ▼
     ┌─────────────────┐
     │  Arduino Mega   │  Data pins D2…D13, D66…D71 → LED bars
     └─────────────────┘
                ▲  GND (common reference)
                │
   ┌────────────┴──────────────┐
   │ 5V, ≥3 A switching PSU    │  ← external PSU
   └───────┬───────────┬───────┘
           │           │
           │ +5V       │ GND
           ▼           ▼
┌─────────────────────────────────┐
│ Grove Ports (VCC rail)          │ <– external 5V injected here
│ 8 × LED Bar cables              │
└─────────────────────────────────┘
by Christian S. Perone
Uncategorized

Arduino WAN, Helium network and cryptographic co-processor

I was recently interested in the intersection of Machine Learning and RF and I was taking a look into LoRa modulation, which is based on Chirp Spread Spectrum (CSS), and ended up getting to know more about the Helium network. I still think that the most stupid piece of technology behind crypto mining is spending GPU/CPU/ASIC cycles to do proof-of-work (PoW), but in the Helium network, they did something quite interesting, which was to switch to something useful such as the proof-of-coverage instead of generating heat and burning energy. Therefore we can say that the miners are doing something useful by providing radio coverage, instead of purely generating heat.

(more…)

by Christian S. Perone
Genetic Algorithms, Pyevolve, Python

The Interactive Robotic Painting Machine !

I’m glad to announce a project created by Benjamin Grosser called “Interactive Robotic Painting Machine“. The machine uses Python and Pyevolve as it’s Genetic Algorithm core, the concept is very interesting:

What I’ve built to consider these questions is an interactive robotic painting machine that uses artificial intelligence to paint its own body of work and to make its own decisions. While doing so, it listens to its environment and considers what it hears as input into the painting process. In the absence of someone or something else making sound in its presence, the machine, like many artists, listens to itself. But when it does hear others, it changes what it does just as we subtly (or not so subtly) are influenced by what others tell us.

Read more about the project in the Benjamin Grosser website.

 

by Christian S. Perone

Tags

announce arduino Article bayesian benford law c cnn convolutional neural networks covid deep learning evolution evolutionary algorithms feature extraction ga genetic algorithm Genetic Algorithms genetic programming hardware Image Processing jit jython karl popper language model llm LLVM machine learning Math matplotlib modis News nlp Philosophy programming Pyevolve Python pytorch raspberry pi Science scikit.learn sigevolution tf-idf twitter ubigraph uncertainty visualization
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.