Back to Projects

AI Video Production System

Overview

A multi-agent AI pipeline for producing educational YouTube content about artificial intelligence. Rather than relying on a single generalist model, the system distributes the work across five specialized agents — each with a focused role, custom persona, and specific output format.

Agent Architecture

Agent Role Output
ResearcherBot Gathers facts, stats, expert opinions on a topic 3–6 sourced bullet points
WriterBot Turns research into a complete 1,400+ word script Full video script with hook/intro/body/CTA
EditorBot Polishes tone, pacing, hook strength Revised script
VisualBot Suggests on-screen elements and B-roll Timestamped visual cues
BrandingBot Adds channel intro, outro, and CTAs Final production-ready script

WriterBot Persona

The WriterBot is tuned to a specific voice: conversational yet sophisticated, narrative-driven, dryly humorous, and research-obsessed. The goal is content that treats viewers as intelligent but not experts — building genuine understanding rather than surface-level summaries.

Key techniques built into the system prompt:

Why Multi-Agent?

A single LLM prompt attempting to research, write, edit, and brand simultaneously tends to produce mediocre work across all dimensions. Specialization — giving each agent a narrow, clear mandate — produces better output at each stage. It also makes the pipeline easier to debug: if the hook is weak, the problem is in WriterBot’s prompt, not in a monolithic black box.

Technology Stack


GitHub Repository ← Back to Projects