Aircraft Maintenance QuoteBot

Aircraft Maintenance QuoteBot

Inspiration

Post pandemic, we have seen a surge in demand for air travel, which airlines from all across the world scrambled to reactivate their entire fleet of aircraft in order to capture this rising demand. The increase need to send their aircraft for routine maintenance and repair, is evident as more customers are requesting us for quotations for their aircraft maintenance.

As an aircraft maintenance repair and overhaul, or MRO service provider, it is paramount for us at Malaysia Airlines to provide the best service for our customers. But the primary task of generating quotations often takes a long time and requires long hours of manual assessments and evaluations. As we already have a massive amount of historical data, it is about time that we create a programme that can automatically generate a quotation by leveraging the tremendous power of large language models (LLMs).

What it does

Our QuoteBote is able to tremendously expedite the assessment of the requested aircraft maintenance tasks and provide a quicker and more efficient way of producing quotations. Besides that, it can also be built upon to offer targeted repair suggestions and other maintenance recommendations to our customers.

How we built it

We built our QuoteBot by integrating multiple solutions within the Microsoft Power Platform (i.e. Power Automate and PowerApps), Databricks and language models (i.e. LangChain and OpenAI LLM) into one seamless process.

Challenges we ran into

While doing the project, we immediately faced the challenge of maneuvering the integration complexity as a result of different environments and tools being used. We also discovered that the model serving feature is unavailable in our region. As such, we had to design a solution that could run the LLM without having to use such feature.

Moreover, as we packaged our solution as an application using Microsoft PowerApps, we quickly realised that the user experience is key to ensure the ease of use and to minimise any potential input errors by users. As a result, user feedback is necessary for this solution to be accepted by our stakeholders and be deemed successful in the long run.

Accomplishments that we're proud of

We are proud to be able to work together as a team across different entities within our organisation, with little to none experience dealing with LLMs. All of us had to learn everything from scratch and in the end, to be able to successfully build a fully functional QuoteBot despite facing numerous challenges along the way, makes all our hard work and perseverance even more worthwhile. These challenges shaped our project journey, marked by resilience, innovation, and a commitment to delivering efficient, user-centric solutions.

What we learned

Seamless Tool Integration Mastery: Gain a deeper understanding of LangChain features like prompts, chains, and agents, unlocking enhanced capabilities in developing dynamic conversational experiences.

Langchain Agent Synergy: A key takeaway is the adept combination of the LangChain agent with OpenAI for handling CSV data. This synergy offers a deeper understanding of how to leverage different technologies for enhanced language processing and data analysis.

User Interface: As we are packaging our solution in an application, the user interface plays an important role in order to achieve a high quality user experience and to minimize any possible occurrence of user input error.

Continuous Improvement: We also learnt that collecting user feedback and implementing continuous model training are necessary to improve the overall system of our QuoteBot, which we will need to carry on in order to ensure the future success of this project.

What's next for Aircraft Maintenance QuoteBot

We envision our QuoteBot to perform web scraping to obtain more detailed information and history of the particular aircraft based on its registration number in order to propose specific maintenance recommendations.