# Clippi | Sherwin Vishesh Jathanna > 🏆 3rd Prize at the Agentic AI Innovation Challenge 2026. Clippi is an AI-powered, node-based video editor you can talk to. Instead of scrubbing timelines, you… Source: https://www.sherwinvishesh.com/project/clippi Copyright 2026 Sherwin Vishesh Jathanna. Text may be quoted with attribution. The design and source code are not licensed for reuse. 🏆 3rd Prize at the Agentic AI Innovation Challenge 2026. Clippi is an AI-powered, node-based video editor you can talk to. Instead of scrubbing timelines, you drag clips onto a canvas and describe your edits in plain English while Mistral AI, Meta's SAM 2, and a vision model handle segmentation, search, and processing behind the scenes. Video editors like Premiere Pro and DaVinci Resolve are incredibly powerful, and also genuinely intimidating, and if you actually watch someone edit, most of their time is not spent being creative, it is spent *scrubbing back and forth* hunting for the right two seconds of footage. **Clippi** throws that workflow out. It is an *AI-powered video editor you talk to*: you drop your clips onto a canvas, wire them together like a flowchart, and describe the edits you want in plain English. Ask it to blur a background or find the product demo, and it figures out what you mean, does the work, and updates the video live. It won **3rd Prize at the Agentic AI Innovation Challenge 2026**. ### An editor shaped like a flowchart Instead of a scrolling timeline, Clippi gives you a *node-based canvas*. Each clip is a node you drag onto the board, and you connect them to build your sequence, where the connections themselves double as transitions like cuts and fades. The frontend is built in **React**, with **Remotion** rendering live previews right in the browser, so the edit you are describing takes shape in front of you in real time rather than after a long export. It turns editing into something closer to sketching a diagram than wrestling with a control panel. ### Just say what you want The intelligence is what makes Clippi *agentic*. **Mistral AI** is the brain: it reads each request, works out your intent, and decides which tools to fire. When you want to isolate a speaker or blur everything behind them, **Meta's SAM 2** handles the object segmentation. When you say *find the part where I open the box*, a vision model that has indexed every frame lets you *search your footage by meaning* and jump straight to that moment, no scrubbing required. And **ElevenLabs** fills in the audio side with AI voiceovers, dubbing, and captions. You are not clicking through menus, you are just telling an editor what you want and watching it happen. ### The heavy lifting lives on a GPU All of that machine-learning work is far too heavy to run in a browser tab, so Clippi splits itself in two. The lightweight frontend stays snappy in the browser, while a **FastAPI** backend running **FFmpeg** and **OpenCV** lives on a *rented cloud GPU* and does the real processing. That separation is what lets Clippi segment objects, index frames, and render effects on genuinely demanding footage without ever freezing up the interface you are working in. ### Keeping the AI honest The piece I am proudest of is a guard against the AI *hallucinating edits it never made*. Every action the model triggers is logged to **Weights & Biases** before it runs, and afterward the model's own account of what it did is checked against that log. If it ever claims to have applied an effect that never actually happened, the system catches the discrepancy and makes the model go back and correct itself. It is a small idea with a big payoff: you can genuinely *trust* that the finished video matches what Clippi says it did. Built together with **Divyam Kataria**, Clippi is, at its core, an editor you can have a real conversation with: drop in your raw footage, describe the video you want, and watch it assemble the finished cut. ## At a glance - **Timeline:** Jul 2026 - **Type:** Hackathon Project - **Event:** Agentic AI Innovation Challenge 2026 - **Award:** 3rd Prize - **Other Contributors:** Divyam Kataria ## Technologies - React (Vite) - Tailwind CSS - Zustand - Xyflow - Remotion - FastAPI - FFmpeg - MoviePy - OpenCV - Mistral AI - Meta SAM 2 - Pixtral Vision - ElevenLabs - Weights & Biases ## Categories - Web App - AI/ML - Hackathon Submission ## Links - [GitHub](https://github.com/sherwinvishesh/Clippi) - [Devpost](https://devpost.com/software/clippi) - [Demo video](https://www.youtube.com/watch?v=hgIqbaxhBpU)