Human Machine Teaming for Rapid Asterix Sensor Integration

Presenting a novel end to end pipeline for turning labor intensive and time consuming data standardization into a scalable and easily labeled workflow.

Comment

Inspiration

We were inspired by the topic proposed by Scarlett from DISA. The general idea being that it is extremely cumbersome to write a parser for an external sensor, and costly for the government to have the company who built the sensor later furnish them a parser to a desired format of choice.

For this reason we built a tool made for decoding a sensor stream with nothing more than an ICD document, which is challenging unstructured data full of images tables and text. Due to the lack of data examples, especially parsed, when doing a task like this, tools like this need to work hand in hand with a human operator for best success.

We'd like to thank the CDAO individual for reviewing our work during the hackathon and suggesting improvements that leveraged humans to verify AI output.

Also thanks to the gentleman that loaned us his RTL SDR.

What it does

At a high level a human pairs with an Advanced Rag Pipeline to quickly parse a ICD and create a message decoder. This is done through a human-machine-teaming GUI.

To show the value in using a tool like this, we quickly stood up a sensor to Common Operational Picture for plane data pipeline. (Custom GUI and Encode/Decode pipeline to Asterix format).

How we built it

We used llama-index, llama-cloud and llama cloud, rtlangle

Challenges we ran into

We had to borrow hardware when we decided to try and do the end to end pipeline

Accomplishments that we're proud of

The parser generated using an advanced rag pipeline actually was able to read ads-b from an ICD, and the end to end looks good with the current air picture.

What we learned

Humans are important for verification with sensitive data, and there aren't always examples that an agent ai can self revise to try and improve the initial decoder with.

What's next for Human Machine Teaming for Rapid Asterix Sensor Integration

Scale, we want to get our system working as soon as possible. Unification to Asterix has been slow enough and our tool will save many working hours.

Built With

Share this project:

Updates