(3)
(2)
(1)

🚀 Delay LLaMA: Your AI Copilot for Predicting Flight Delays

✈️ What inspired us

Our team wanted to tackle a real-world problem that millions of travelers face daily: flight delays. Inspired by our own experiences waiting anxiously at airports, and seeing the economic + emotional impact delays cause, we thought:

"Why not build an AI ‘LLaMA’ detective that inspects the skies and helps predict delays in advance?"

Combining the pun of a “LLaMA” (like an LLM model) with an actual cartoon llama looking for clues on a runway gave our project a fun, memorable twist!

🛠️ How we built it

🧩 Frontend: A clean, responsive UI built with Streamlit, allowing users to input weather conditions (temperature, wind speed, visibility) and instantly see predictions.
🔮 Backend & ML:
- Powered by Google’s Gemini API (Generative AI LLM) to perform playful explanations of why a flight might be delayed.
- Used a custom-trained XGBoost regressor for predicting the minutes of delay based on historical weather and flight data.
🔗 API: Hosted on Render.com with a FastAPI server to serve predictions.
🐙 Version Control & Collaboration: Managed on GitHub, enabling all team members to contribute seamlessly.

🔍 What we learned

How to effectively combine traditional ML (XGBoost) with Generative AI (Gemini LLM) to deliver both predictive analytics and engaging natural language outputs.
Deploying ML + GenAI workflows on limited resources under tight hackathon timelines.
Improving our UX by making even technical delay predictions friendly and fun, so anyone could use the tool without needing a data science background.

🧗 Challenges we ran into

Setting up a seamless pipeline from training locally to deploying on Render.
Managing API latency and cold starts, especially with free tier limits.
Integrating Gemini’s LLM with structured outputs so we could feed predictions to it in a conversational way.
Designing the playful brand (Delay LLaMA) while ensuring it still conveyed professional value.

💡 What's next for Delay LLaMA?

We’d love to expand Delay LLaMA to:

Use real-time weather + flight feeds (like OpenSky & NOAA APIs).
Incorporate more advanced GenAI features to explain probabilistic causes of delays.
Offer insights to airlines for dynamic scheduling or maintenance optimization.

🎉 Closing thoughts

This project was an exciting blend of machine learning, generative AI, and creative branding that brought smiles to our team (and hopefully, to future travelers).
We’re proud that Delay LLaMA helps turn stressful travel moments into something a little more predictable — and a lot more delightful.

✅ Made with ❤️ at our MLH hackathon, by a team of developers passionate about making tech both useful and joyful.