AI Air Assistance

• 

  travelers in panic mode

Inspiration

Seeing the recent Air India flight 171 crash in India due to some fault where more than 200 people lost their life with only one survival, created AI powered voice assistance system.

What it does

It will collect plane vibration/ sensor data to predict probability of crash. Based on the probability score, it will trigger survival instruction associate with the nature (local condition) of the emergency. Voice assistance will be provided to all based on a prompt like “fire in back portion of the plane”.

How we built it

We used AWS services like Amazon generative AI services like Bedrock for taking input text and generating phrases that can be converted to audio. These audios can be pushed to the edge device where the travelers can listen and follow the survival instructions.

Challenges we ran into

Triggering the emergency based on the sensor data was very complicated. Once we finalized the triggering point, we setup the AWS services for AWS Bedrock and Amazon Polly. Passing the text as well as image prompt for the model to understand and generate the survival instructions accordingly. The generated output also takes into consideration the traveler age, set location and their language.

Accomplishments that we're proud of

Based on the emergency trigger, the survival instruction is passed on to the travelers based on the ages and in their language. All are done autonomously whenever trigger happens.

What we learned

All the AWS services mentioned here and their configuration. Used Claude for the text and image prompt together.

What's next for AI Air Assistance

Make it fully autonomous using MLOps. Will wrap this application with different tools from MLOps. All the data can be collected from sensor and stored in the AWS S3 bucket. The emergency trigger model can be finetuned based on the new dataset collected whenever there is model drift. Finally, it can be made into container using Docker and pushed to AWS ECR creating endpoint for the travelers in the flight.

Built With

• aws-cloud
• fastapi
• json
• python
• sagemaker
• scikit-learn