Skip to content
KunanonJ

Portfolio · AI agents

AI Automation

A team-built set of internal AI agents at The Binary Holdings. The main one, Fando, finds sports clips, has Gemini judge and caption them, queues them for human approval and publishes the approved ones.

All projects

AI agents for content ops and hiring.

What it does

  • Clip sourcing from YouTube, Instagram, TikTok, Twitch and X
  • Gemini checks that each clip is real match content and writes its title and caption
  • Duplicate detection with perceptual image hashing
  • A Google Sheet as the human approval queue
  • Approved clips are trimmed, stored in Cloud Storage and published through a Playwright browser flow
  • A recruitment agent that scores resumes, grades take-home cases and sends interview invites, with human review for borderline candidates

The problem

Finding, checking and re-uploading short clips was manual work, one clip at a time. The agents take over the repetitive steps and leave the judgement call, approve or reject, to a person.

How it is built

Python agents on Google's Agent Development Kit with Gemini models, deployable to Vertex AI Agent Engine. Media handling uses yt-dlp and ffmpeg, and the recruitment agent is a FastAPI service. CI runs ruff, mypy and pytest on GitHub Actions.