Self-Generating Brand Orchestration Engine

Content platforms are excellent at capturing output but fall short when it comes to preserving judgment. Marketing teams waste valuable hours repeatedly making identical decisions because each tool isolates the reasoning behind previous choices. This results in:

• Lost context across campaigns and customer personas
• Repetitive manual generation of "new" content
• No record of decision-making rationale

The Solution - Self-Generative Orchestration

A tool that remembers why decisions were made and how they were executed, preserving your brand's institutional knowledge. It captures reasoning through Graph RAG (Relational Memory) and executes it via deterministic DAG orchestration. The LLM simply writes content-it doesn't control the process.

In essence:Your CoWorker that never forgets how you think.

How It Works

1. Flow Chaining - Natural Language → DSL → Action DAG

• Tasks are parsed into a slot-based DSL, then recombined using existing flows
• Adapter Scoring ranks which flows best fit together
• Example: "Search comments → Generate reply template" connects two existing flows with no new logic required

2. Graph RAG - Relational Memory

• Goes beyond simple vector similarity
• Creates connections between Trends, Drafts, Posts, and Comments in a unified knowledge graph
• Sidecar Watcher synchronises embeddings and edges every 30s–5min in real time
• The system remembers not just what happened, but why it was chosen

Architecture Overview

The system combines sophisticated components across multiple layers:

• Backend: FastAPI + Celery + PostgreSQL (pgvector) + Redis
• Orchestrator: Deterministic DAG Executor, Slot-based DSL
• Graph RAG: Vector + Graph Search with 9 node types × 7 edge types
• Workers: CoWorker/Sniffer/Synchro autonomous background agents
• Adapters: Threads + Instagram (Meta Graph API)
• Frontend: React 19 + TypeScript + Vite + Zustand + shadcn/ui
• Interface: Chat-based orchestration with card-driven UX
• Visualisation: Interactive Graph Explorer and real-time KPI dashboards

Practical Use Cases

Two-Stage Publishing

"Get trends → Create draft → Schedule → Auto-publish" Creates deterministic, replayable campaign execution.

Reactive Automation (5-Minute Loop)

"Keyword → Tag → Auto DM / Comment" Persona policies ensure safe, contextual replies at scale.

Comment-Driven Creation

"Comment clusters → Response template" Behaves like a brand new feature, built entirely by recombining existing flows.

Technical Strengths

• DAG-based Orchestrator with idempotent action nodes
• Graph RAG Sidecar for real-time vector + graph synchronisation
• Playbook Intelligence logging every Persona × Campaign decision
• Deterministic Automation that's reproducible, debuggable, and safe
• CoWorker Agents that run your brand routines even when you're offline

Business Value

• Minimal Code Growth - new features emerge through recombination, not expansion
• Deterministic Behaviour - reproducible DAGs simplify debugging
• LLM Cost Efficiency - rules handle routing while LLM handles only final phrasing
• Full Auditability - every Persona, KPI, and Flow Chain is logged
• Predictable Scalability - 0% code growth, 100% reuse

Architectural Philosophy

"AI shouldn't replace human judgment - it should remember it."

The app is built on the principle that recombination is cheaper than addition. Instead of shipping new code for every feature, it connects existing flows like LEGO blocks through adapters.

Summary

It's not merely a content tool-it's an orchestration system that captures why decisions happen, not just what they produce. Every past action becomes future intelligence, allowing your marketing team to build on previous successes rather than reinventing the wheel.