Climate mate

Climate Mate is an AI application (RAG) that utilizes reliable data from sources such as the IPCC and NASA to educate users about climate change and empowers individuals to take action.

Comment

Inspiration

The motivation behind our collaborative project stems from a desire to leverage modern technologies, such as large language models and Google AI services, to create a climate mate application. Drawing inspiration from initiatives like the Climate Fresk workshop, which explains the causes and impacts of climate change, we aim to develop an application that educates users about climate change and empowers them to take action while providing an enhanced user experience.

What it does

We've made this app that's all about answering your questions on climate change in a way that's super clear and spot-on. We used some fancy stuff like semantic search and prompt engineering to make sure you get the best, most relevant answers possible. We utilised Gemini that we used throughout the app to make sure the answers are top-notch and the whole experience is smooth and easy for you. So, basically, you ask a question, and our app gives you exactly what you need to know about climate change, no fuss.

How we built it

The application itself is a microservice built with Golang. The solution is running on GKE. The AI part built to recognise the user query, create embeddings and generate answers utilises Google AI (Gemini). PostgreSQL with pgvector extension was picked to store the indexed data and supported us with the semantic search. We have hand-picked pdf reports from very trusted sources like IPCC and NASA.

Challenges we ran into

Balancing semantic search to consider all data sources. Tuning the prompts to better fit the output expectations.

Accomplishments that we're proud of

It seems to generate very little hallucinations and provides precise answers based on indexed data and the user query.

What we learned

Working with RAG

What's next for Climate Mate

We plan to expand our document collection, tune our prompts and semantic search methods; add re-ranking based on ragas or similar framework.

Built With

Share this project:

Updates