Skip to content
$ jader.ms hire_me()
back to projects
case study · self-built · 2025 — present

Project Delta Multi-profile AI content production — source to ready-to-publish video.

Most "AI content tools" stop at writing captions. Delta runs the full chain: picks a source from the queue, writes the script with Claude, narrates it with ElevenLabs, renders the video with per-scene media + text overlay, and publishes to every connected platform. Each profile is fully independent — its own brand voice, rules, schedule, accounts.

01 · Profiles overview — multi-profile dashboard with per-account stats and auto-schedule countdown
02 · Strategy editor — 8 tabs per profile, enforcement rules applied at generate time
03 · Playground draft — paste a source, generate scenes, render, promote to a real profile
8
strategy tabs per profile
4
platforms (TikTok / IG / YT / Linktree)
5
pipeline stages source → script → narration → render → publish
click from source URL to ready-to-publish video
01/ Pipeline

From source to publish

┌──────────┐    ┌──────────┐    ┌──────────────┐    ┌──────────┐    ┌──────────┐
│ SOURCE   │ →  │ SCRIPT   │ →  │ NARRATION    │ →  │ RENDER   │ →  │ PUBLISH  │
│          │    │          │    │              │    │          │    │          │
│ products │    │ Claude   │    │ ElevenLabs   │    │ FFmpeg + │    │ TikTok   │
│ topics   │    │ + system │    │ voice +      │    │ scenes + │    │ Instagram│
│ manual   │    │ prompt + │    │ multilingual │    │ overlay  │    │ YouTube  │
│ Reddit   │    │ segments │    │ + audio mix  │    │ + Ken    │    │ Linktree │
│ articles │    │          │    │              │    │ Burns    │    │          │
└──────────┘    └──────────┘    └──────────────┘    └──────────┘    └──────────┘
     │               │                  │                 │               │
     ▼               ▼                  ▼                 ▼               ▼
  queue           enforcement      audio             video           history
  used-flag       rules pass       file              file            + retries

  ───── workflow gates can pause between any two stages for approval ─────
02/ Capabilities

What it does

Multi-Profile, Multi-Platform

Each profile owns N social accounts across TikTok, Instagram, YouTube, and Linktree — with its own brand voice, schedule, and connections. Scale from one account to dozens without forking the system.

8-Tab Strategy Editor

About · Rules · Source · Sound · Look · Schedule · Workflow · Connections. Every part of how a profile behaves — from forbidden words to narrator voice to text overlay font — is editable per-profile.

End-to-End Generation Pipeline

Source → script (Claude) → narration (ElevenLabs) → video render → publish. Workflow gates can pause for approval between stages. Or skip the gates and let auto-mode run the full chain.

Template System with Timed Segments

Each template defines a system prompt + segments with second-counts (Hook 3s · Setup 12s · Escalation 30s · Climax 25s · Cliffhanger 15s). Reuse a winning format across profiles.

Playground Sandbox

Validate ideas without committing to a profile. Pick any template, paste a Reddit post or article, generate, render. Promote what works to a real profile in one click.

Enforcement Rules at Generate Time

Forbidden words, required CTA, caption length cap, emoji policy, hashtag rules (prefix + niche pool with min/max count), case enforcement — applied by the pipeline before anything ships.

Visual System per Profile

Brand palette presets, custom color tokens (BG / text / accent / secondary), primary + bold fonts, text overlay styling. Per-scene media chains with stock footage fallback (Pexels) and Ken Burns motion.

Per-Post History + Error Tracking

Every published post recorded with full caption, post ID, and timestamp. Failed publishes keep the response headers + body so you can debug instead of guessing.

03/ Stack

What it's built with

Backend
Python FastAPI SQLAlchemy SQLite Pydantic
AI
Anthropic SDK (script generation) ElevenLabs SDK (narration) Multi-voice / multi-language System-prompt + segment composition
Media
FFmpeg pipeline Per-scene media fallback chain Ken Burns motion Text overlay rendering Pexels stock fallback
Integrations
TikTok API Instagram Graph API YouTube Data API Linktree OAuth flows + session persistence
Frontend
JS dashboard Multi-tab strategy editor Live pipeline status Playground / draft editor
$ next.step()

Need a custom content pipeline?

Multi-platform, multi-profile, AI-native. I design and ship pipelines that turn raw sources into ready-to-publish content — with the rules, voice, and visual identity you want.

Get in touch
Next case study
Personal Agent — Cross-device AI command center