
A dynamic LED strip display integrated behind my wall-mounted surfboard that visualizes real-time tide and wave conditions. 🏄‍♂️🌊

![Surfboard LED Wall Mount](/images/work/show/surfboard_led_wall.jpg)

The display changes color based on the tide level - transitioning from yellow at low tide to deep blue at high tide. When the waves are firing and conditions are perfect, the LEDs blink to let me know it's time to grab my board!

- Raspberry Pi with LED strip control
- Real-time tide and wave data integration
- Color gradient mapping for tide levels
- Blink animation for optimal surf conditions

![LED Display in Action](/images/work/show/surfboard_led_display.jpg)

This project combines my passion for surfing with electronics, creating an ambient display that's both functional and aesthetically pleasing. The LEDs are hidden behind the surfboard mount, creating a subtle glow effect that doubles as room lighting while keeping me updated on surf conditions at a glance.


Surfboard Tide & Wave LED Display


<img
  src="https://github.com/erwannbst/imtoundra/assets/16354899/58aae6c8-b9eb-4a58-ba8d-2208d1498d1e"
  alt="IMToundra banner"
  title="IMToundra banner"
  />

IMToundra is a project that aims to monitor the environment and detect animals in the Toundra house. It consists of two units: the first unit is located inside the Toundra House and is responsible for detecting animals and monitoring the environment. The second unit is located outside the Toundra house and is responsible for sending the data to the cloud.

## Flowchart

<img
  src="https://github.com/erwannbst/imtoundra/assets/16354899/977da74f-3836-42ef-9b3d-d24325303846"
  alt="IMToundra schema"
  title="IMToundra schema"
  />

#### 1. Animal detection
***This step is not implemented in this project.***
The first unit detects an animal and takes a picture of it.

#### 2. Image recognition
The image is sent to the image recognition microservice, which detects the animals in the image.

#### 3. Sensor reading
Then the unit reads the sensors (humidity, temperature, and CO2 level) and sends the data to the encoding microservice.

#### 4. Encoding
The encoding microservice encodes the data to reduce the payload size and sends it to the network microservice.

#### 5. Network (loraspberry)
The network microservice sends the data to the second unit using LoRa.

#### 6. Database
The second unit receives the data and sends it to the database microservice.


## Hardware

This is a list of the hardware used in the project:

#### First unit

This unit is located inside the Toundra house and is responsible for detecting animals and monitoring the environment.
It has no internet access and sends the data to the second unit using LoRa.

- Raspberry Pi 3B
- ENS160 humidity sensor + AHT21 temperature sensor
<img
  src="https://github.com/erwannbst/imtoundra/assets/16354899/07a51419-0ccc-47a3-a6e7-425202ede7c4"
  alt="ENS160"
  title="ENS160"
  style="display: inline-block; width: 250px;"
  />

- LED for visual feedback
- Ra-02 LoRa module (SX1278)
<img
  src="https://github.com/erwannbst/imtoundra/assets/16354899/89a290d7-6e87-47ea-ad91-12d788398494"
  alt="SX1278"
  title="SX1278"
  style="display: inline-block; width: 250px;"
  />


#### Second unit

This second unit is located outside the Toundra house and is responsible for sending the data to the cloud.
It has internet access and uses Wi-Fi to send the data to the cloud.

- Raspberry Pi Zero W
- Ra-02 LoRa module (SX1278)


<img
  src="https://github.com/erwannbst/imtoundra/assets/16354899/12b72760-f707-435e-a4b3-8423cff42c1e"
  alt="IMToundra schema"
  title="IMToundra schema"
  />

## Software

Several microservices to detect animals and send analysis to a database.

- Image recognition
- Encoding
- Network
- Database
- Analytics

### Image recognition

This microservice utilizes the YOLOv7-Tiny model to detect objects in an image and return the results in JSON format.

The `recognition_visual.py` file displays the image after detection and serves as a visual example for the microservice located in the `recognition_ms.py` file.

#### Features

- The microservice exposes an `detect_objects` endpoint that accepts POST requests with images.
- It loads the pre-trained YOLOv7-Tiny model as well as classes from files.
- Detections are performed on the provided input image.
- Only detections with confidence above 0.5 are retained.
- Redundant detections are removed using Non-Maximum Suppression (NMS).
- The results are returned in JSON format containing detected classes and their confidences.

##### Usage

1. Send a POST request with an image to the `/detect_objects` endpoint.
2. The microservice returns detected object detections in the image.

##### Execution

- Run the Python script to launch the microservice.
- The microservice runs on `http://localhost:5000` by default.

### Sensors

Microservice of the fist unit with the sensors. It is responsible for reading the sensors and sending the data to the encoding microservice.

### Encoding
Its job is to minimize the weight of the payload
it removes the keys from the json and encodes the values to bytes with `msgpack`.
```
curl -X POST -H "Content-Type: application/json"
-d '{"animal_identifie": "cat", "timestamp": 1641992400, "temperature": 25, "co2": 400}'
http://localhost:5000/encode
```

returns
`{
  "encoded_bytes": "94a3636174ce61ded0d019cd0190"
}`

### Network (loraspberry)
This microservice is responsible for sending the data to the second unit using LoRa.


### Database
Build the container
`docker build -t imtoundra-db .`

Run the database container
`docker run -d -p 5432:5432 imtoundra-db`



IMToundra


# Train for your next JobInterview with AI-powered recruiters !

JobInterview AI is a mobile application that helps you prepare for your next job interview. It uses OpenAI's GPT-3 to simulate a job interview and give you feedback on your answers.

<img
  src="/images/work/show/jobinterview-screen1.webp"
  alt="JobInterview AI"
  title="JobInterview AI"
  style="display: inline-block; width: 250px"
/>
<img
  src="/images/work/show/jobinterview-screen2.webp"
  alt="JobInterview AI"
  title="JobInterview AI"
  style="display: inline-block; width: 250px"
/>

[Link to the App Store](https://apps.apple.com/cm/app/ai-job-interview-trainer/id6471625704)

JobInterview AI


# Mistral Client for iOS

I wanted to use the French Mistral AI model on my phone, but there was no mobile app available. So, I decided to build my own iOS client using React Native and Expo. This project allowed me to create a custom interface for interacting with the Mistral AI model on my mobile device. To handle the backend operations securely, I implemented a Node.js server, ensuring a robust and efficient communication between the app and the AI model.

## Features

- Custom UI for Mistral AI interactions
- iOS compatibility
- Built with React Native and Expo for rapid development


## Technical Details

This project was built using React Native and Expo, allowing for quick development and easy deployment to iOS. The app interfaces with the Mistral AI API to provide a seamless experience for users who want to interact with the AI model on their mobile devices.

## Security
To ensure the security of the Mistral AI API key, I implemented a server-side proxy using Node.js and Express, deployed on Render.com. This approach prevents exposing the API key in the client-side code. The proxy server handles all API calls to Mistral AI, securely storing the API key as an environment variable. I used HMAC for request authentication between the iOS app and the proxy server, ensuring that only authorized requests are processed. This security approach effectively protects the API key while allowing secure interactions with Mistral AI.


<img
  src="/images/work/show/mistral1.jpg"
  alt="Mistral Client Screenshot 1"
  title="Mistral Client Screenshot 1"
  style="display: inline-block; width: 250px"
/>

<img
  src="/images/work/show/mistral2.jpg"
  alt="Mistral Client Screenshot 2"
  title="Mistral Client Screenshot 2"
  style="display: inline-block; width: 250px"
/>

Mistral Client




A minimum viable product order to create a calendar-based app to better organize team events

- Firebase backend
- Android & iOS support with React Native
![FullWeek](/images/work/show/fullweek-screenshot.png)

![FullWeek](/images/work/show/fullweek-screenshot-animation.png)

---

Links:
- [Landing webpage](https://fullweek.fr)

Full Week


# Discover best places to fly

Mobile app available on iOS and Android to share best places to fly your drone

- [Check this app](https://dronespot.fr/mobile)

![DroneSpot Landscape](/images/work/show/dronespot-landscape.png)

A web app is available too ! [https://dronespot.fr](https://dronespot.fr)

Drone Spot


Starting with a small school project with a friend, we ended up building a real-time two player game. The whole app is written in JavaScript with Node.js and Socket.io. We used the awesome platform Glitch to develop and host the app, you can check it out [here](https://age-of-ampecqueur.glitch.me/).

![Game](/images/work/show/game.png)

Game


I rent a garage in a building with a door controlled by a radio remote. I wanted to be able to open the door from my phone.
I first started by looking at the remote and the receiver. I wanted to copy the signal from the remote but I figured out that it was a rolling code system so it was not possible that easy.

![GarageRemote](/images/work/show/garage_remote.jpg)

I decided to keep the original emitter and to control it with a Raspberry. The Raspberry would receive the instruction from my phone and would trigger the emitter through an optocoupler.

![RemoteCircuit](/images/work/show/remote_circuit.png)

This is the circuit I ended up with.

I then added a Siri shortcut so that I can now open the door without even touching my phone.

Garage remote


![Drone](/images/work/show/drone_overview.jpg)

As I was at the end of my high school, I wanted to learn about drones by building one. It was a lot of fun to build and to fly. I had to learn a lot about electronics and programming to make it work. The software is based on Betaflight.

![Drone](/images/work/show/drone_zoom.jpg)
The electronic part was the most challenging. I had to learn about ESC, motors, batteries, and a lot of other things. I also had to solder and properly use a multimeter, before having start my electronic studies.


![Drone](/images/work/show/drone_final.jpg)

I then took part in [a competition with the association Anjou Drone Racer](https://www.ouest-france.fr/pays-de-la-loire/maine-et-loire/angers-pilotes-de-drones-ils-realisent-le-reve-d-icare-680a2c4c-dca4-11e9-8deb-0cc47a644868). Unfortunately, a soldering went wrong, it burnt and crashed. I still have to fix it.

Thanks to another pilot, I have been able to participate with its backup drone and ended up 9th over 14, here is [the leaderboard](https://chronomatch.fr/event/qualifrecap.php?event=186) 






Drone


A small dashboard beside my desk to quickly check the surf reports and the weather forecast in the nearest surf spots. 🏄‍♂️

![SurfReport](/images/work/show/surfreport_side.jpg)

- Raspberry Pi 3
- Python script to scrap the data from a website

![SurfReport](/images/work/show/surfreport_front.jpg)

Here I can see that the waves are not that good and I'd better stay working on some projects.