PostKonnect

The problem

I built it for small teams where one person looks after social media on top of their actual job. Some of those teams photograph children, and that changes how every picture has to be handled.

What I built

PostKonnect drafts with Gemini, and the drafting turned out to be the easy part. Most of my effort went into approval and consent. Every draft waits for a person to approve it, and every image carries a record of who agreed to its use.

What it does

• One input becomes a draft for Instagram, Facebook, LinkedIn, TikTok and WhatsApp.
• An approval queue where drafts can be edited before they are scheduled or sent.
• Face detection in the browser, so children’s faces can be blurred before anything is uploaded.
• Consent records linked to the photographs they cover.
• A brand voice that every draft reads before it is written.
• A phone mode for platforms that still have to be posted to by hand.

Under the hood

• Next.js and TypeScript on Supabase.
• MediaPipe face detection runs in the browser, so photographs are blurred before they leave the device.
• Gemini model names are looked up at runtime, after Google retired the old ones twice in a month.
• Posts go out through Make, with a safeguard against sending the same post twice.

Where it stands

It is live and used daily. I made the face detection deliberately cautious, since a missed face costs a child their privacy and a wrong box only costs someone a tap to remove.

Why it might matter to you

If you’d like AI in your work but want a person to approve what goes out, this is how I’d approach it.