Merging Passions: From GenAI to Conscious Travel Planning ??

I've recently joined Loka , an AI powerhouse that helps companies build and launch projects. One of the first goals that I set out to do was to attend the LLM Zoomcamp, an online course about real-life applications of LLMs. And, that's exactly what I did.

For our final project, we were tasked with developing a practical application using Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG). In this article, I will share how I chose what project to develop and what I ended up implementing in the end.

I hope this can serve as an inspiration for people joining the next cohorts of the course!

How a personal challenge became my motivation

As an ML engineer working in generative AI solutions for biotech companies, my day-to-day work focuses on developing projects that can help improve patient outcomes and develop novel therapeutics through digital innovations. It's a field I'm deeply passionate about: using data and technology to solve healthcare problems has been guiding my career since day 1.

However, when the time came to choose a final project for the course, I decided to explore another great passion of mine: travel. ??

In the last few years, I've been pushing for my trips to be as conscious and ethical as possible. I really believe in the power of responsible tourism and how it can shape and contribute to local communities (thank you Shivya Nath for the learnings). Whether it's choosing locally run restaurants, supporting community-driven tours, or staying in eco-friendly accommodations, I've been making an extra effort to take more conscious decisions. However, planning such trips, especially in the era of mass tourism, can be challenging.

By the time the final project of the course came, I was planning my summer holidays in Albania (btw, amazing country, totally recommend!). And then again, I was facing this exact challenge, so I thought: what if I could create a tool to help travellers like me make more sustainable and ethical choices?

My research led me to an intriguing project and published article about a similar concept – A Green Destination Recommender – but it was limited to European cities. As my love for Southeast Asia (SEA) has grown over the years, and I will be planning a big trip there next year, I realised there was an opportunity to create something valuable for this region, which is particularly affected by the impacts of mass tourism.

And so TravelSEA Advisor is born

TravelSEA Advisor was the project that I decided to implement: a RAG-powered system providing personalised, sustainability-focused travel recommendations for Southeast Asian countries. My main goals were:

1. Customised Trip Planning: generating tailored itineraries based on user preferences like budget and travel style.

2. Sustainability Focus: highlighting eco-friendly accommodations and experiences, encouraging responsible tourism.

3. Southeast Asia Expertise: specialise in countries like Thailand, Vietnam, Cambodia, and more – regions particularly vulnerable to over-tourism and climate change impacts.

Technical Challenge

For the ones interested in the technicalities, here is a concise summary of what I ended up implementing:

- Data Sources: I leveraged WikiVoyage content on both Southeast Asia countries and sustainability travel content.

- RAG Implementation: I experimented with various retrieval approaches, including MinSearch, Chroma DB, and PGVector, which ended up being the one chosen for the final app. For the RAG system, I used OpenAI GPT4o model and a personalised prompt message in the system.

- Evaluation: I evaluated the different retrieval approaches by synthetically generating a ground truth dataset using an LLM. I then used this dataset to evaluate the retrieved documents using two metrics: Hit Rate and Mean Reciprocal Rank.

- User Interface: I developed a simple Streamlit-based UI for easy interaction.

- Containerisation: I implemented Docker for improved deployment and scalability.

Small disclaimer: This was what I was able to implement on the period of time I had available, but I know a lot more can be done to improve the solution in order to make it ready for a real-world. I mention some of the possible improvements in the project's Readme file.

If you want to check out the full project on GitHub, it is publicly available here: TravelSEA Advisor Repository. All feedback is welcomed and appreciated!

P.S. 1.0: I want to thank to the team responsible to put together the LLM Zoomcamp course, specially Alexey Grigorev , for creating this valuable, free, open-source learning opportunity for so many people. Your efforts have significantly contributed to my learning journey, as well as of many other in the community. ??

P.S. 2.0: I'm also very grateful to Loka , and particularly Telmo Felgueira , for supporting my participation in this course and providing the resources needed to develop this project. Loka 's commitment to employee growth is outstanding!! ??

P.S. 3.0: Finally, I'm thrilled to share that this project was successful, earning me the certificate for the LLM Zoomcamp course ??

#AI #SustainableTravel #LLMs #MachineLearning #TravelTech #HealthcareAI