ShortPitch AI

Inspiration

I've always enjoyed watching game highlights, but I often wished for more control over what I saw—whether it was my favorite team, specific players, or key moments like home runs or incredible saves. That’s why I built ShortPitch, an AI-powered video assembler that curates personalized highlight reels based on MLB game data and existing highlight footage. By offering tailored highlights, ShortPitch enhances fan engagement and preserves memorable moments in an entirely new way.

What It Does

ShortPitch generates personalized 2-minute highlight videos based on user preferences. Fans can select specific teams, players, or events to focus on, ensuring they get the content that matters most to them. By integrating MLB data with Google Gemini’s text generation capabilities and Eleven Labs’ natural voice synthesis, ShortPitch provides lifelike commentary that enhances the viewing experience, making highlights feel more dynamic and engaging.

How I Built It

ShortPitch leverages MLB game data and Google Gemini’s AI capabilities, including video understanding and tool calling, to generate structured video instructions. These instructions are then processed within a React-based web application, using Remotion to render high-quality, interactive video content.

The pipeline works as follows:

MLB data + AI Processing: Gemini analyzes MLB data and generates structured video instructions. Text To Speech + Cloud Storage : Audio commentaries are generated from the video instructions and stored on Google Cloud Storage.

VideoClipping : Using Gemini's Video understanding feature, only the relevant sections of the videos are shown and the rest is cut of.

LLM to Video Conversion: The web app parses these instructions into a video data representation. Rendering with Remotion: The parsed data is fed into Remotion components to produce a seamless, AI-generated highlight video.

Challenges I Faced

Working with evolving APIs and the unpredictability of LLMs posed challenges, particularly in ensuring stable video generation. At times, the web parser crashed due to incomplete data or incorrect format.

However, through rigorous testing and iteration, I developed a system that generates high-quality highlight reels with minimal issues.

Accomplishments I am Proud Of

Successfully integrating two of my favourite tools Google Gemini and Remotion library for react, pushing the limits of both technologies for myself.

Creating a smooth AI-powered highlight generation system that personalizes sports content at scale. Learning a lot more about MLB —it was an exciting journey!