TrendPilot

Video prompt and plan
Video upload scheduling
My channel analytics
Trend data
Video content

Inspiration

Every creator hears the same advice: post consistently, 2-3 times a day, to grow. But nobody talks about the hidden cost behind that advice the hours spent researching what's trending, checking what's coming up on the calendar, and figuring out what actually worked last time, all before a single script gets written. That repetitive research loop is the real bottleneck, not creativity. We wanted to build something that handles that loop automatically, so creators can spend their time actually making things instead of researching what to make.

What it does

TrendPilot pulls real trending topics, upcoming festivals and cultural moments, and a channel's own past performance data, then uses AI to generate a ranked 7-day content plan complete with the reasoning behind each pick. For every planned video, it generates a full script, a ready-to-use AI video generation prompt, and SEO-optimized titles, descriptions, and tags. The generated video prompt hands off directly into a video generation tool, and the finished video slots back into the plan as ready to schedule.

How we built it

The AI planning layer runs on Google's Gemini API this is fully live in our demo, generating a real plan and real scripts on every run, not templated output. Trend and calendar data comes from free, official sources (YouTube trending data structure, public holiday APIs). Since live YouTube account integration requires OAuth from the viewer's own channel, we built the app with a clearly-labeled DEMO_MODE that runs on realistic sample YouTube data for trending videos and channel analytics, while the YouTube upload, scheduling, and analytics-fetching code is fully implemented and ready to run live once connected to a real account. The dashboard is built in Streamlit for fast iteration, with the backend in Python.

Challenges we ran into

The biggest challenge was scoping honestly under a tight deadline: it would have been easy to fake a "fully live" demo, but we decided transparency mattered more than appearances so we built real, working OAuth and upload code, but gated it behind DEMO_MODE and said so clearly, rather than pretending live YouTube integration was demo-ready when judges couldn't actually test it against their own accounts in a short video. Getting the AI ranker to return consistently well-reasoned, structured plans (not just generic suggestions) also took real prompt iteration.

Accomplishments that we're proud of

We're proud that the core AI reasoning combining trends, calendar context, and channel history into a ranked plan actually works, live, end to end. We're also proud of building something we'd genuinely use ourselves: this isn't a hackathon toy, it's a tool that solves a real weekly headache for anyone trying to grow a channel.

What we learned

We learned how much of "automation" tools is really about good data orchestration pulling the right signals together rather than the AI step itself, which is almost the easy part once the inputs are good. We also learned the value of being upfront about what's live versus simulated in a demo; it's a better story than pretending everything is finished, and it's more useful feedback for judges too.

What's next for TrendPilot

Next is flipping DEMO_MODE off for real: connecting live YouTube OAuth so creators can plan against their actual channel, enabling real scheduled uploads via the YouTube Data API, and closing the feedback loop pulling in real analytics after each post so next week's plan adapts based on what actually performed, not just what was predicted to.