Inspiration-
I was peacefully scrolling through YouTube Shorts when suddenly—BOOM—a cliffhanger. A faceless narrator, a half-told fact about black holes, or a skit ending in “To be continued…” I searched. I scrolled. I cursed. I wasted 15 minutes of my life I’ll never get back trying to find Part 2. Then, my teammate yelled: *WHERE IS PART 2? * And just like that, a star was born (the project, not us).
What it does-
*Live Creator Analysis: * Real-time YouTube data so you're always current. *AI-Powered Matching: * Trained on 790+ Shorts for peak detective energy. *Smart Learning: * Gets better as people scream “this isn’t part 2!”. *Complete Series Detection: * Like Netflix autoplay but with actual intelligence. *Always Up to Date: * No stale Shorts here. *Beautiful UI: * We didn’t sleep, but we made it pretty.
AI, But Make It Spicy-
The semantic search matches Shorts even if titles or thumbnails are wildly unhelpful (e.g., “Part 2???” with 19 emojis). Our model clusters videos by title similarity, creator matching, time gaps, and deep vibes (okay, embeddings). Plus, it learns from negative feedback. *Rage-clicking helps! * (Because we don't have enough datasets)
How we built it-
*Frontend: * React 18 + TypeScript + Vite. *Styling: * Tailwind CSS + Framer Motion. (read: smooth as butter) *Database: * Supa base. (PostgreSQL flavor) *AI/ML: * Hugging Face Transformers via Xenova. (because we like pain) *APIs: * YouTube Data API v3. *Graphics: * WebGL (OGL) + Custom Shaders. *Built with: * Bolt – AI-powered, emotion-draining IDE.
Challenges we ran into-
Xenova errors (we don’t want to talk about it) Bolt duplicating files like it’s playing a prank. Supa base limits. Money limits. API limits. Brain limits. Making a demo video. ** _Turns out we’re developers, not actors. Who knew? _ **Bolt Limits It can't handle large project AI integration asks for money but we don't have any. _ Funds- _Resources=Money
Accomplishments that we're proud of-
Making something that actually solves a daily annoyance. ** *Learning AIML from scratch. (aka hours of crying and Googling) * **Sticking to it despite DNS struggles, ML bugs, and “where the hell is this part coming from” moments. *Built a complete working app, deployed via Netlify, with a custom domain and AI that kind of slaps. *
What we learned
DNS is not “Do Not Stress” (spoiler: you will). ** **Machine Learning ≠ Magic – More like, “Math crying in a trench coat pretending to be AI.” Bolt has a sense of humor: ** _it kept duplicating our components like it was cloning dinosaurs. _ *Bolt is far better for making prototypes. * **Netlify has trust issues – It took 48 hours and still didn’t believe we updated our DNS. Gaslight, gatekeep, Netlify. Console.log is our therapist – _The only one who listens without judgment. _ Git is a time machine – _Until your git reset --hard into a black hole and lose your will to live. _
What's next for WhereIsPart2
Instagram & TikTok Integrations- Because what’s the point of finding the “part 2” if you can’t immediately blast it into the algorithmic void for clout? Smarter AI (a.k.a. fewer breakdowns) We’re planning better AI integration — ideally one that doesn’t forget what a “part 2” is halfway through a video. _ Feeding it Big Data- Like, really big. So big it makes ChatGPT look like it’s doing kindergarten math. All in the name of “accuracy,” of course. Social Media Bots- Why search when a bot can do it for you faster than your brain can finish saying “wait, what was the username again?” Monetization (_Because our servers need snacks too) We'll soon be introducing a payment gateway, so users pay X rupees/dollars/hopes/dreams to access our gloriously trained model. Pay to search… capitalistic, but in a cool tech-startup way. (Bonus) Maybe a feature to find Part 1 too Because sometimes even the beginning is missing, and we like symmetry.
Built With
- css
- framer
- ide
- motion
- react
- tailwind
- typescript
- vite
- webgl
- xenova
- youtubeapi

Log in or sign up for Devpost to join the conversation.