Inspiration
As the number of short-form content creators continues to grow, the amount of information shared through platforms like Instagram Reels and TikTok is also increasing rapidly.
However, videos often do not clearly provide details such as product names, locations, or related links, and it can be difficult to tell whether the claims made in a video are actually true. In these cases, users have to search through comments, look up information themselves, and compare multiple sources to verify it.
SignalSkip is designed to reduce that friction. It helps users find the information they need and check the evidence behind claims without interrupting their short-form viewing experience.
What It Does
SignalSkip provides two core features.
The first is information extraction. It analyzes speech, text, objects, and music in a video to identify useful information such as product names, locations, and related details.
The second is fact-checking. It extracts key claims from the video and searches public sources to determine whether there is evidence supporting those claims.
The main goal of SignalSkip is to reduce the time and effort users spend manually searching for information and comparing multiple sources.
How We Built It
SignalSkip is built as a multimodal pipeline combining speech transcription, OCR, object detection, audio analysis, public-source retrieval, and content fingerprinting.
We use NVIDIA Nemotron as a verification layer that compares extracted claims with retrieved evidence.
Tiger Data stores analysis results, evidence, confidence scores, and content fingerprints, allowing us to reuse previously analyzed and verified information.
ElevenLabs converts verified results into short spoken explanations, while DigitalOcean provides the infrastructure for deployment and processing workloads.
Challenges
Short-form videos are noisy. Scenes change quickly, text can be small, and speech, music, and visual information can all appear at the same time.
Another challenge is that OCR, speech, and visual analysis can produce conflicting results. Instead of forcing a single answer, SignalSkip preserves these conflicts, adjusts confidence when necessary, and can abstain when there is not enough evidence.
What We Learned
Through this project, we learned that reducing the effort required to find and verify information can be just as important as generating the information itself.
SignalSkip is not designed to make decisions for users. Instead, it helps them quickly find the information they are looking for and understand the evidence behind it.
Ultimately, SignalSkip aims to reduce the manual searching required to answer questions like, “What was that video talking about?” or “Is this information actually true?”
Log in or sign up for Devpost to join the conversation.