AVIS — AI-Powered Video Intelligence System

The AI copilot for live broadcast directors.

The Problem

Live event broadcasting requires a skilled director to constantly monitor multiple camera feeds and make split-second switching decisions, choosing which angle goes live at any given moment. This demands years of experience, intense focus, and is mentally exhausting over long events. During a 3-hour sports broadcast, a director may make hundreds of switching decisions, and fatigue inevitably leads to missed moments.

Smaller events like school sports, community concerts, local gatherings simply can't afford a professional broadcast director. The result is either a poor single-camera stream or no broadcast at all.

Our Solution

AVIS is an AI copilot that assists broadcast directors in real time. It continuously analyzes every camera feed using the Presaige API, scores each angle on engagement, visual quality, and content relevance, and either automatically selects the best camera or provides ranked recommendations for a human director to act on.

AVIS doesn't replace the director, it augments them. Like autopilot for broadcasting: it can fly solo when there's no crew, and assist when there is one.

How It Works

Pipeline

Multiple Camera Feeds
        ↓
  Upload clips to Presaige API
        ↓
  Presaige scores each clip on multiple metrics
        ↓
  AVIS ranks cameras and selects the best
        ↓
  Viewer sees a continuous, high-quality broadcast

Scoring with Presaige

Presaige's API provides deep content analysis that goes far beyond simple image quality. Each clip is scored on seven. These scores allow AVIS to make nuanced decisions — not just "which camera is pointed at the action" but "which angle will captivate the audience most."

Streaming Architecture

  • The camera streams are scored using clips from every few seconds, keeping rankings fresh
  • FFmpeg handles real-time remuxing from MP4 to MPEG Transport Stream, a packetized format designed for seamless concatenation (the same format used in broadcast TV)

The Platform

AVIS serves three different users with tailored interfaces:

Viewer View

The audience sees a single continuous stream showing the best camera angle at all times. There's no buffering or glitches during camera switches. A minimal overlay shows which camera is active and a countdown to the next potential switch. Viewers never need to know multiple cameras exist, it just looks like great coverage.

Director View

The broadcast director gets full control and visibility:

  • Grid view of all camera feeds simultaneously
  • Real-time Presaige scores overlaid on each feed
  • Ranked sidebar showing cameras ordered by score
  • One-click override to force any camera live
  • Auto/Manual mode toggle to let AVIS drive or take manual control
  • History timeline showing which camera was selected in past intervals

Camera Operator View

AVIS also helps every camera get better. Using Presaige's analysis, each camera operator receives actionable recommendations in real time. This creates a feedback loop: better individual camera feeds lead to better overall broadcast quality.

Beyond Broadcasting

AVIS isn't just for sports and events. The same technology unlocks value anywhere multiple cameras exist.

Wildlife & Zoo Cameras — Dozens of cameras across a zoo or wildlife reserve. AVIS auto-curates a highlight stream showing the most interesting animal activity. Visitors see lions playing, elephants bathing — not empty enclosures. Researchers get auto-filtered footage instead of reviewing hours of empty frames.

Security & Surveillance — Security teams monitor 50+ feeds, which is humanly impossible. AVIS flags the most interesting feed at any moment — unusual activity, movement, anomalies — reducing cognitive load and helping catch incidents faster.

Concerts & Music Festivals — Smaller acts can't afford full production crews. AVIS turns a few fixed cameras into a professional-looking livestream for fans at home.

Esports & Gaming — Auto-switch between player cams, overhead views, and action replays. Highlight the most intense moments across multiple player perspectives at tournaments of any size.

Tech Stack

Layer Technology
AI Scoring Presaige API (engagement & quality metrics)
Backend Python, FastAPI, async scoring pipeline
Streaming FFmpeg (MPEG-TS remuxing), real-time proxy
Frontend Next.js, React, mpegts.js for stream playback

Data Sources:

  1. APIDIS dataset. Image and Signal Processing Group (UCLouvain). (2022, October 13). https://ispgroup.gitlab.io/code/apidis/#data
  2. Tsuchida, S., Fukayama, S., Hamasaki, M., & Goto, M. (2019, November). AIST Dance Video Database: Multi-genre, Multi-dancer, and Multi-camera Database for Dance Information Processing. Proceedings of the 20th International Society for Music Information Retrieval Conference, ISMIR 2019. Delft, Netherlands. https://aistdancedb.ongaaccel.jp/

Built With

Share this project:

Updates