DeepDive

DeepDive turns the travel inspiration you save on TikTok and Instagram into something you can actually use. You paste a reel or link; and AI pulls out places, vibes, activities, and destination hints.

Comment

Inspiration

Planning trips is time-consuming and often overwhelming. We already spend a lot of time on our phones looking at travel content, so we wanted a way to turn that inspiration into an actual trip—automatically and without the hassle.

What it does

Users can paste in social media links, and our AI analyzes the video by extracting transcripts, captions, and locations. It then generates travel recommendations based on vibe, location, and user interests.

How we built it

Backend: Built with FastAPI, with routes for extraction, trip generation, revisions, planning from posts, async reel processing, and TikTok/Gemini-style parsing.

Media & AI: We use yt-dlp to download videos when needed. Gemini analyzes video and text to suggest locations and extract signals. A processing pipeline normalizes this data and resolves locations using Google Places, including support for Google Maps share links.

Frontend: A map-focused single-page app (DeepDive.html) built with React (via CDN + Babel) that communicates with the backend using config.js. We also explored a Next.js app with optional Supabase integration for authentication and data storage.

Tooling: Makefile scripts handle setup and config generation. Pytest is used to validate extraction behavior.

Challenges we ran into

API access was a major issue—we didn’t anticipate how long Instagram and TikTok API approval would take. Because of this, we pivoted to web scraping and direct links, which limited some of the data we could access and likely reduced result quality.

We also faced challenges configuring the extraction pipeline and ensuring consistent, accurate outputs from the AI processing stages.

Connecting endpoints to the database and maintaining valid, secure data flow was another hurdle.

Finally, integrating the frontend and backend smoothly took significant time and effort.

Accomplishments that we're proud of

We’re proud of our ability to pivot quickly from using official APIs to building a scraping-based solution that still delivered results.

We also worked well as a team and were able to complete the project without major issues or conflicts.

What we learned

We learned that there’s always another way to solve a problem. This project also gave us experience building a full end-to-end system, including working with LLM APIs, backend development, databases, and frontend integration.

What's next for DeepDive

We plan to improve the UI and overall user experience. We also want to integrate official Instagram and TikTok APIs, especially to allow users to pull from saved posts directly, which would make the product much more seamless.

Share this project:

Updates