Challenge
The system had to connect browser collection, local video processing, transcription and several AI roles without letting agents overwrite each other's work, invent sources or drift from shared formats.
AI / CONTENT AUTOMATION / RESEARCH PIPELINE
From raw Reels research to a finished script through a traceable AI pipeline
Content Factory is a file-based multi-agent system for producing Instagram Reels. It collects source videos and visible data, extracts frames and audio, creates transcripts, identifies patterns, updates a playbook, generates ideas and moves each script through repeated critique and optimization.
- Python 3
- Claude Opus
- Anthropic Messages API
- Groq Whisper
- MLX Whisper
- FFmpeg
- Markdown / JSON
- Instagram / TrendSee
Why the system was created
Why the system was created
Content work quickly becomes a maze of links, notes, hypotheses and disconnected script versions. Content Factory creates a repeatable path where every conclusion has an input file, every stage has a dedicated role and the final script preserves the complete research and revision chain.
Who the system serves
The platform serves content teams, experts and agencies that publish Reels regularly and want to work from observed examples, their audience, services, offers and calls to action rather than starting every script from a blank page.
Solution
We built a Python pipeline with independent agent contracts and file exchange through Markdown and JSON. Codex handles collection and raw-media processing, Claude Opus performs analysis and content creation, and Critic with Optimizer iteratively improve each script until it reaches the configured quality gate.
Capabilities
Product capabilities
The capabilities support one operating scenario and share common state, control and analytics.
Reels source collection
The system accepts a video manifest, local media, captions, visible metrics and selected comments.
Video decomposition
FFmpeg extracts audio and frames, while a dedicated tool can assemble local video from discovered media assets.
Automatic transcription
Audio is transcribed through Groq Whisper or local MLX Whisper, with a backfill path for missing transcripts.
Diagnosis and patterns
AI turns the raw packet into a trend report, viral diagnosis and recurring patterns for later stages.
Playbook and ideas
New evidence updates the shared knowledge base before a strategist creates ideas from audience, services, offers and CTAs.
Critique and optimization
A draft receives independent evaluation and successive rewrites until it passes the configured quality gate.
Outcome
Outcome
Content Factory turns Reels research and scriptwriting into a repeatable production process. The team receives more than a standalone AI-generated text: it gets a traceable chain from source video and pattern to idea, critique, improvements and final script.
Python 3
Claude Opus
Anthropic Messages API
Groq Whisper
MLX Whisper
FFmpeg
Markdown / JSON
Instagram / TrendSee
New project
Working on a similar task?
Describe the current situation and the outcome you need. We will understand the context and propose a practical first stage.