During my internship at CAE, a global leader in flight simulation and training that manages crew rostering, I noticed a significant business problem: the manual process of reading and validating crew rosters against the rules set out by the DGCA (Directorate General of Civil Aviation). Crew schedulers often had to sift through lengthy documents to ensure compliance with flight regulations, leading to inefficiencies and potential errors. To address this, I developed an AI-powered chatbot that allows users to verify crew assignments and clarify regulations directly from PDF documents.
The chatbot interacts with complex documents like flight regulations, reports, and CRM user guides, making it easier to retrieve specific information and verify decisions. As a demonstration, I trained the chatbot to understand DGCA's civil aviation requirements on 'FLIGHT CREW STANDARDS TRAINING & LICENSING'. Now, users can ask it questions such as:
"The time difference between Singapore and India is 2 and a half hours. If a crew member flies from India to Singapore, do they need to be acclimatized?"
"Can you show all rules that are related to consecutive nights?"
"Is it okay for a crew member to fly 80 hours in a 14-day consecutive window?"
These queries can be answered quickly and accurately without manually going through the entire document. You can test out this prototype by navigating to https://flight-reg-assistant.streamlit.app.
TECHNICAL DETAILS
The RAG was created using "BAAI/bge-small-en-v1.5" as the embedding model, FAISS as the vector store, and Llama 3.1 powered by Groq as the conversational model