Projects / PennApps XXVI

Award winnerCATEGORY - Privacy / SecurityConfidence: low

Superhuman

A swarm of AI agents searches CCTV footage in parallel, restores damaged frames and matches suspects from text descriptions.

Original submission ↗View source ↗Watch video ↗Use as a reference →
Project facts
Event
PennApps XXVI
20–21 Sep 2025
Team
2 people
Build
software only
AI
multi agent, computer vision, generative media

Problem

Investigators review surveillance footage clip by clip, too slowly for urgent cases.

Solution

Over twenty agents analyse feeds in parallel; video inpainting restores occluded frames; vision-language models and face re-identification match natural-language descriptions; a generative layer predicts trajectories.

Demo moment

Team-described: describe a person and see matching clips across cameras.

Why it stood out

  • organizer statedWon “CATEGORY - Privacy / Security” among 85 submissions.Source ↗
  • team statedStacks many techniques (inpainting, re-identification, VLM retrieval) in one pipeline.Source ↗
  • curator inferenceWon the privacy and security category while being a surveillance tool; the civil-liberties trade-off is not addressed in its summary.

Lessons for your next hackathon

  • Technical depth can win a security category, but expect hard questions on misuse.

Similar situations

Projects that share domains, mechanisms or judging lenses — often with a different problem and stack.Tune the comparison.

Sources

  1. https://devpost.com/software/superhuman · devpost project · retrieved 2026-10-13
  2. https://pennapps-xxvi.devpost.com/project-gallery · devpost gallery · retrieved 2026-10-13