-
-
Turn real audience comments into grounded themes, reply-worthy signals and next-content opportunities.
-
Catch missing sponsor requirements with grounded evidence, timestamps, exact tokens and targeted fixes.
-
Inspect the actual MP4 before it goes live: format, resolution, audio, pacing, opening and CTA.
-
PublishSignal closes the creator loop: check content before publishing, then learn from real audience feedback.
Inspiration
Creators already have plenty of tools that help them make more content. But publishing has another problem: how do you know the content is actually ready to go live? And after publishing, how do you turn audience reactions into something useful for the next piece of content?
That is the gap we wanted to solve with PublishSignal.
Instead of being another content generator, PublishSignal focuses on the quality and feedback loop around publishing:
Check before you publish. Learn after you do.
What it does
PublishSignal gives creators one workflow for both pre-publish quality control and post-publish audience learning.
1. Short-Form Preflight
Creators can upload an actual MP4 for TikTok, YouTube Shorts, or Instagram Reels.
PublishSignal checks:
- orientation and aspect ratio
- resolution
- duration
- audio presence
- speech/activity timing
- long low-energy gaps and pacing
- opening/hook quality
- call-to-action quality
The important distinction is that this is not only an LLM prompt. PublishSignal actually inspects the media file and combines deterministic media checks with AI-assisted semantic analysis.
The result is a readiness score, grounded findings, and prioritized fixes before the video goes live.
2. Sponsored Content / SponsorGuard
Sponsored campaigns add another layer of risk: missing a promo code, mentioning the wrong URL, forgetting a required phrase, making a prohibited claim, or saying something too late.
PublishSignal turns campaign requirements into explicit checks and evaluates them against an SRT transcript.
It supports checks such as:
- required mentions
- exact required tokens
- timing constraints
- required URLs
- forbidden phrases and claims
Every finding can include timestamped evidence from the original transcript.
When something is missing, PublishSignal can also generate a targeted correction. For example, if CREATOR25 is required but absent, it can recommend:
Use code CREATOR25 at checkout.
The exact protected token remains unchanged.
3. Audience Pulse
Publishing is not the end of the workflow.
Audience Pulse helps creators understand what their audience is actually telling them after a post goes live.
It can work with public YouTube / Shorts comments when the YouTube API is configured, or with manually pasted comments from TikTok, Instagram, stream clips, and other platforms.
Audience Pulse surfaces:
- positive reactions
- questions
- content requests
- constructive criticism
- confusion
- low-information comments
- recurring themes
- comments worth replying to
- grounded next-content opportunities
Instead of asking AI to invent generic content ideas, the suggestions are grounded in real comment IDs and real audience feedback.
How we built it
PublishSignal is a full-stack web application.
Frontend
- React
- TypeScript
- Vite
- custom English / Spanish localization
- light, dark, and system themes
- local draft persistence
- responsive editorial UI
Backend
- Python
- FastAPI
- Pydantic
- PyAV for real MP4/audio analysis
- deterministic compliance and media-analysis services
- structured provider boundaries for AI features
AI
We use Gemini for bounded semantic tasks such as:
- sponsor brief requirement extraction
- opening and CTA analysis
- improvement suggestions
- audience classification
- grounded themes and next-content opportunities
AI outputs are validated against strict schemas and grounded back to original input wherever possible.
Production
- Frontend deployed on Vercel
- FastAPI backend deployed on Render using Docker
- backend-only API secrets
- production CORS
- rate limiting
- health checks
- safe error handling
The project also has more than 800 automated backend and frontend tests covering parsing, media checks, compliance, grounding, persistence, localization, error handling, and production behavior.
Challenges we ran into
One of the biggest challenges was deciding what should use AI and what should remain deterministic.
For example, checking whether a required coupon code exists should not depend on an LLM. That belongs in a deterministic rule engine. But judging whether an opening clearly communicates a topic or identifying repeated audience themes benefits from semantic analysis.
Another major challenge was provider reliability. During development we hit real Gemini rate limits, so we designed PublishSignal to fail gracefully.
If semantic analysis is unavailable:
- deterministic checks still survive
- uploaded or pasted source data is preserved
- Audience Pulse does not invent fake percentages
- unavailable sections are clearly marked as not evaluated
- users can retry without losing their work
That resilience became an important part of the product.
Accomplishments we're proud of
We are proud that PublishSignal became a working creator workflow rather than a single AI demo.
During the hackathon we shipped:
- real MP4 analysis for short-form content
- deterministic sponsor compliance
- timestamped transcript evidence
- targeted fix generation
- post-publish audience intelligence
- grounded next-content opportunities
- English and Spanish UX
- responsive light/dark themes
- production deployment
- extensive automated testing
Most importantly, the three modules work together as one loop:
Create → Check → Publish → Listen → Learn → Create better content
What we learned
We learned that AI becomes much more useful when it is constrained by product rules, validated structures, and real source evidence.
We also learned that creator tooling does not always need to generate more content. Sometimes the more valuable question is:
Is this ready to publish?
And after it goes live:
What is my audience telling me to do next?
What's next
We would like to expand PublishSignal with:
- deeper platform integrations
- creator review history and comparisons
- richer post-performance signals
- improved audience trend analysis
- more short-form platform presets
- additional sponsor campaign rule types
- creator/team collaboration workflows
The long-term vision is a creator operations layer that helps creators make better publishing decisions before and after every post.
Try PublishSignal
Live app: https://publishsignal.vercel.app
Source code: https://github.com/rmogrovejo/sponsorguard-ai
Log in or sign up for Devpost to join the conversation.