Masthead

The problem

I wanted to maintain a useful professional voice without beginning every post from a blank page. Alerts could surface news, a general chatbot could produce copy, and a design tool could make slides, but the research trail, editorial judgment, and publishing calendar still lived in separate places.

Masthead made that a single, reviewable operation: automation prepared source-led work while the named author kept approval control.

What I built

The system joined five stages that are usually separate:

1. Discovery. It searched for recent material against a brand’s content pillars and extracted the underlying sources.
2. Triage. It classified and scored each item, recorded the reasoning and suggested angle, and separated approved signals from the rest.
3. Planning and drafting. Approved sources could feed a content plan or become reactive LinkedIn drafts written against the brand’s strategy and voice.
4. Visual production. It generated brand-aware, square carousel layouts and exported them as PDFs suitable for LinkedIn document posts.
5. Review and distribution. A user could edit, reject, schedule, or publish a draft, including its carousel, rather than allowing the pipeline to post by itself.

Interactive product walkthrough

From source signal to scheduled post

Reconstructed from the product before work ceased. These are illustrative interface states: no live data is loaded and no action is sent to a backend. Choose a stage, or use Previous and Next, to follow the workflow.

MMasthead

Pipeline ready

JN

Step 1 of 5: Fetch. 3 example signals prepared.

Factory / Sources

Signal inbox

Example data

Regulation84

Industry update from a monitored source

Recent · source extracted · analysis ready

Markets76

Market commentary matching a content pillar

Recent · source extracted · analysis ready

Transactions71

Transaction announcement from a public source

Recent · source extracted · analysis ready

System prepared

Search results, source text, classification, and relevance score.

Human decision

Choose the signal worth pursuing.

Record retained

Source, rationale, pillar, and angle.

← Previous01 / 05Next →

The operating product

I built Masthead as a React and Firebase application rather than a collection of disconnected scripts. Its dashboard included a signal inbox, content-plan view, draft editor, generation controls, and a drag-and-drop publishing calendar. The backend handled ingestion jobs, model calls, usage metering, carousel generation, and LinkedIn publishing.

The product grew beyond my own account. Organisations could contain multiple brands with separate pipelines, members, and permissions. I added onboarding, plan limits, usage reporting, administrative controls, and Stripe subscription infrastructure. That work demonstrated the scope required to turn a personal workflow into multi-tenant software; it did not demonstrate that people would pay for it.

Where judgment stayed human

Masthead could find an article and propose what to say about it, but it could not decide whether the point was worth attaching to someone’s professional identity. Approval remained a product boundary: content quality, copyright, attribution, and brand risk still required human review.

Why I stopped

The system worked well enough to prove that it could be built, and I took it far beyond a prototype. But its one committed user was me. I had not established that enough professionals or firms shared the problem, wanted the entire workflow rather than a smaller combination of tools, or would pay enough to support the ingestion, generation, and review costs.

Continuing to add features would have made the implementation more complete without answering the business question. I ceased work rather than mistake more implementation for evidence of a market.