Project Info

Klip AI

Devpost

Inspiration

Early-stage tech founders and builders know their product deeply, but most do not have the time, budget, or knowledge to advertise. As sudents and builders, we faced this issue ourselves; we coded many personal projects but didn't know where to find users. In this era, social media is one of the best methods for promotion. Klip AI started from one question: can a founder turn a website/application URL into social media reels in minutes?

What it does

Klip AI converts a startup’s public website URL into multiple short-form reels in under 10 minutes. Flow: Founder pastes a URL. Our agent scrapes all pages in the website and extracts product messaging and key visuals. Our agent does deep research to understand the context behind the product - market research, competitor analysis, and other information about the website on the internet. It also researches patterns in trending reels on social media so it can emulate those styles. The agent generates one reel based on this research context. The user has a session with an AI Marketing Director avatar which will ask questions about the user's style preferences for the reels. The agent creates a plan for 3-5 more reels and generates them with this additional information. Founder exports MP4 files and posts. Speed-to-first-reel: No editing timeline, no design tools, no blank page.

How we built it

Scraping + capture: Browserbase StageHand for page discovery, screenshots, and text extraction. Trend analysis: Engagement-weighted template scoring from current short-form format signals. AI layer: Perplexity Sonar-based deep-researched content and LLM-generated hook/copy options mapped to product features and audience intent. Video pipeline: HeyGen Video Generation API, Remotion + Google Veo for programmatic composition, transitions, and rendering. Backend: FastAPI orchestration, Node Frontend: React

Challenges we ran into

It was difficult for HeyGen's realtime Avatar API to converse when there was background noise, because of its auto VAD. So, we decided to optimize the auto VAD by implementing Kalman filters in the frontend to remove background noise before passing it into the avatar stream. It was difficult to make the videos include accurate images of the website because LLMs/image models often hallucinate. So we used BrowserBase's Stagehand to dynamically take the best screenshots and images in the website when scraping it, then passed them into HeyGen. Video generation is expensive. We parallelized jobs and generated reels before-and-after the session with the Marketing Director so users can see reels quickly. Accomplishments we’re proud of End-to-end pipeline from URL to downloadable reels. Sub 6-minute generation for multiple reels. We created reels for the founders and hackathon participants and posted them on Instagram: https://www.instagram.com/klip.techfounders?igsh=MzRlODBiNWFlZA==. Being able to have a realtime/low-latency speech conversation with the AI avatar, which asks high-quality questions. The content and visuals in the reels are accurate - no hallucinations.

What we learned

We learned that generating many reels is better than generating a single reel because users prefer multiple variations that they can post and A/B test quickly. We gained technical skills in creating agents that do multi-turn conversations, web automation, video editing, research, and putting that all together in a pipeline that includes both sequential and parallel steps.

What's next

Multi-platform exports (Shorts, Reels, TikTok formats). Deeper personalization from prior performance history. Allowing editing

Analysis

Compare with all teams

View

Metric

Figures cover GitHub contributors during the hackathon window. A co-authored commit counts in full for each author, so per-member totals add up to more than the whole-team figures.

Technology

Found in codeClaimed only
  • AnthropicIn code
  • CSSIn code
  • ExpressIn code
  • FastAPIIn code
  • FirebaseIn code
  • HTMLIn code
  • JavaScriptIn code
  • OpenAIIn code
  • PythonIn code
  • ReactIn code
  • TypeScriptIn code
  • Node.jsClaimed

11 of 12 appear in the indexed code. 1 claimed on Devpost could not be matched to code, which may simply mean the tool leaves no trace in the repository.

AI coding agents

No AI coding agent signals were found in this repository.

Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.

Codebase size

Source size

346 KB

Source files

74

Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.

0 stars