[AI Automation · Business Operations · Product Workflows]
Marketing Brain — AI Operations Workflows
AI-assisted workflows for marketing operations: content planning, task decomposition, project context and reporting, delivered through Telegram-based interfaces.
01Project overview
Marketing Brain applies the same principles as the AI assistant to business operations: structured workflows where AI does the language-heavy work and the system keeps context, structure and history.
These are connected modules of the marketing product rather than separate client projects.
02Engineering challenge
AI is good at drafts and decomposition, but without structure and context its output stays generic. The work here is in the workflow around the model.
- A
Break down broad goals into structured, actionable tasks.
- B
Keep project context and reusable instructions between sessions.
- C
Connect plans with data from analytics.
03Solution architecture
Hover or tap each stage.
Goal
A broad marketing objective from the user.
04Main components
- 01
Content planning
AI-assisted planning of content around goals and channels.
- 02
Task decomposition
Breaking goals into structured, ordered tasks.
- 03
Context store
Project context and reusable instructions kept between sessions.
- 04
Reporting
Summaries combining plans and available data.
05Technology stack
AI
- LLM API
- Structured outputs
Backend
- Python
- Context storage
Interface
- Telegram
06Interactive demonstration
Goal decomposition
IllustrativeIllustrative simulation — scripted output, no model is called.
Goal
Project context used
Structured plan
- Press “Decompose” — the plan appears as validated, ordered tasks.
07Engineering decisions
- 01
Structure over prose
The model returns tasks as data, which can be stored, validated and tracked.
- 02
Context is a first-class object
Project context is stored and reused, not retyped each time.
08Reliability & testing
- Schema validation of AI-generated plans.
- Human review before plans become tasks.
09Code example
from dataclasses import dataclass
@dataclass
class Task:
title: str
channel: str
order: int
def parse_plan(raw: list[dict]) -> list[Task]:
tasks = [Task(str(t["title"]).strip(), str(t["channel"]), int(t["order"])) for t in raw]
if not tasks:
raise ValueError("empty plan")
if len({t.order for t in tasks}) != len(tasks):
raise ValueError("duplicate task order")
return sorted(tasks, key=lambda t: t.order)Illustrative Illustrative Engineering Example — validating an AI-proposed plan.
Production source is private. Examples here show the engineering approach, not proprietary logic. More in the Engineering Lab.
10Possible extensions
- Feedback loop from analytics results into planning.
- Templates for recurring campaign types.