Riddle solving using Gemini AI

What it does

Give an riddle with some options, the model can provide with the correct answer out of the options provided and also can provide with the detailed explanation of the solution.

How we built it

We used Google AI studio to access the Gemini Flash model and have fine tuned it on our riddles data.

Challenges We Ran Into

We ran into some issues using Vertex AI and its rendering functionalities. We attempted to incorporate image classification using fine-tuning of the Vertex AI's models however, it would require hours for the model to be trained. This led to a setback in our project's progress as we were not able to implement the feature of providing then solving riddles based on a possible input 'genre' from an image to the bot.

Accomplishments that we're proud of

Experience in Google AI studio, Google Vertex AI and introduction to Gemini in such a short span is what we are proud of. We are certainly carry forwarding this enthusiasm to implement them in our personal projects.

What we learned

Our team entered the hackathon with little to no experience using Google Vertex AI and Gemini. During our hacking phase, we were able to work fine-tuning the AIs and training the models according to our datasets.

What's next for Riddle solving using Gemini AI

We would like to extend this project such that the model can be capable of accepting images or screenshots of the riddles and analyze it to provide a solution. Basically, we would like to perform image analysis and extract the riddle from the image we provided to the model and then process it to generate correct answer.

Built With

• googleai
• python