Inspiration
Planning a journey or a trip is usually very time-consuming, we need to manually search for various places and determine the route. Further accounting factors like routes, modes, and best places is very time consuming.
With LLMs & AI, we can automate this process to generate the travel paths much faster and easily.
What it does
Now planning can be as simple as: I want to go from X -> Y and visit A, B, C places in order or Find me the best italian restaurants nearby X or Find me details about X place and it'll give you all reviews, address, numbers etc. No need to do manual searches anymore.
Further people, can also ask questions easily relevant to their trip, like "nearby attractions", "how many schools there are nearby X vs Y"
How we built it
- Used Gemini API for understanding the user's intention and segregate into various categories.
- For each category, perform different pipeline, gemini helps in extracting the relevant information from query and we pass it into the google api's functions.
- We use Google places api to get information of places & directions api for making the routes.
- Then Combine all of the extracted information and depending on the usecase, run extra LLM layers, before returning the final output.
Challenges we ran into
- It took time to determine how can we do different things efficiently depending on the user prompt. The workflow of understanding the intention and then running multiple steps took time to implement.
- We couldn't use the gemini-1.5-pro model because we make a lot of calls to API, and we were constantly getting ratelimited.
Accomplishments that we're proud of
- Creating a tool, which any lay person can use, and can improve their daily lives.
- Coding up LLMs + Tools integration from scratch without using any 3rd party library.
- Integrating multiple workflows at once. Saving hours worth of time :)
What we learned
- Gemini APIs
- Understanding Context, and then doing different actions -- Integration of Tools & LLMs
- Google Maps, Places APIs, about latitudes, longitudes, and how to integrate multiple APIs together
- Hacking a MVP in few days :)
What's next for MapsGPT
- Subscribe to Gemini's API plan, and use the 1.5 pro model
- Improve the prompting strategies to also chat back to user if details are needed
- Implement Voice Chat, all user needs to do is just speak :)
- Optimize workflows.
Log in or sign up for Devpost to join the conversation.