Hype It Up

Inspiration

TikTok has already become a hub for social movements and grassroots campaigns. Through TikTok For Good, the platform supports these efforts with tools like Donation Stickers, permanent profile donation links, and the TikTok Change Makers program. While impactful, these features place much of the burden on creators to craft persuasive, trend-savvy content and to align with established NGOs. TikTok’s own guidance emphasizes “proving impact, using specifics, letting creators go off-script, and shouting out supporters.” But not every creator has the skill, reach, or organizational backing to meet those expectations.

Often, TikTok is used informally by everyday people raising funds for urgent needs—like medical bills, surgeries, or basic necessities. These individuals may not be tied to official charities, yet their causes are no less deserving of support. The focus should be on amplifying genuine causes themselves, rather than forcing creators to worry about whether their video is engaging enough to stay relevant in the algorithm.

What it does

Features:

Lets fans boost worthy videos with either Paid Hype (coins → extra reach windows) or Community Hype (free milestones → time-boxed visibility).
Runs a local AI verification (browser-only) that checks consistency (title/desc/ocr/category) and reliability (HTTPS + allow-listed fundraiser domains).
Adds a Context Card, AI summary, suggested hashtags, and a Donate externally button.
Shows Impact UI: projected reach, milestones, and a Hype ledger (visibility events) to motivate.

Tech Stack

Frontend: React + TypeScript + Vite, Tailwind CSS.

Local Browser AI: Transformers.js QA (distilbert-squad) + Summarization (distilbart-cnn-6-6), OCR via tesseract.js.

Architecture: Modular UI (Feed, Creator, Boost, Context Card) + pluggable verifier (localVerifier.ts)

Challenges we ran into

Time Constraints
Switching from a Native application to Web
Noisy OCR (URLs/digits) hurting similarity scores during AI verification
Balancing strictness vs. false rejects (tuned thresholds and added rule fallbacks)

Accomplishments that we're proud of

Fully serverless verification with clear failure reasons (no black-box “AI says no”.)
Fairly clean UI with focused actions despite time crunch
Figuring out how to use HuggingFace for the first time