Apollo

Inspiration

Large Language Model (LLM) to translate sensor data standards described in Interface Control Documents (ICDs) into a schema compatible with the EUROCONTROL Surveillance Information Exchange (ASTERIX) standard for computational decoding.

What it does

Radars collect information about aircraft including its position, its speed, altitude, the type of aircraft etc. In the Air Traffic Center the track of the aircraft is rebuilt, and is displayed on a screen facing the controller to guide the planes. To transmit radar data, ATC systems use an open-source protocol to transport the radar informations called ASTERIX developed and maintained by EUROCONTROL. The major problem with the original specifications is that they are provided in a form of free text (PDF files). As a consequence, the very first step in every asterix project is to retype the specifications to a parsable form like JSON. Our project Apollo translates ASTERIX PDF documentation into parsable JSON format.

How we built it

Apollo is built using (i) an OCR library to convert the PDF into text format and (ii) using LLMs and agents (Radiant, OpenAI, Langchain) for the JSON schema generation.

Challenges we ran into

The main challenge was to get a quick sense of the data input and output sets. We initially focused on finding ASTERIX message data, which ended up not being the the core focus of the challenge as we were not asked to build an end-to-end decoder.

Accomplishments that we're proud of

The Apollo JSON schema generator outperforms the current state-of-the-art JSON project in the space (https://zoranbosnjak.github.io/asterix-specs/).

What we learned

The project forced us to get familiarised with the unstructured data puzzle faced by users of Interface Control Documents (ICDs) along the example of the ASTERIX schemata.