Social Assembly AI Coach for Creators

The AI coach that reviews your content before you post then helps you plan, publish, and grow.

Comment

Inspiration

Every creator we know posts blind. You spend hours on a video, hit publish, and then just… hope. The feedback loop is broken — by the time the analytics arrive, it's too late to change anything. Professional creators fix this with a team: an editor, a strategist, a manager. Most people can't afford one. Our founding creator was doing all three jobs herself at 1 a.m. Social Assembly puts that whole team in every creator's pocket.

What it does

Social Assembly is an AI coaching platform for TikTok and Instagram creators:

  • Pre-post review — the feature people gasp at. Upload a draft clip or photo and get a clear verdict plus specific, frame-aware feedback before it goes live, not after it flops.
  • A coach that knows you — it remembers your niche, audience, and goals, and hands off to specialist agents for monetization (earnings estimates and brand-deal / UGC / affiliate strategy) and content review.
  • Voice-first onboarding — instead of a cold form, "Emma" talks to you in her own cloned voice and interviews you; we synthesize the answers into a living coaching profile.
  • Discovery — a weekly-refreshed feed of trends and paid opportunities matched to your niche.
  • Connected analytics + publishing — link your accounts, feed real follower / engagement / reach numbers into the coach, plan on a calendar, and publish or schedule straight from the app.

How we built it

The reasoning brain is OpenAI (GPT): a CopilotKit chat orchestrator that coordinates two Python A2A specialist agents — a Monetization Advisor and a Pre-Post Video Coach — over the A2A + ag-ui streaming event protocol, so several agents feel like one coach. We route the right model to the right job: OpenAI for language, reasoning, planning, and strategy; Google Gemini for multimodal vision — analyzing the actual pixels and frames of a photo or video — and ElevenLabs for Emma's cloned voice.

The app is Next.js 15 / React 19 (a mobile-first SPA, TypeScript, Tailwind v4, shadcn/ui). Supabase handles auth and a Postgres database locked down with row-level security, so every creator's data is fully isolated. Zernio unifies Instagram + TikTok behind one API for connect, analytics, and publishing, with an HMAC-signed webhook that has no unsigned code path. Monetization runs on Solana Pay (USDC) and Paystack with server-side plan gating. It ships on Render as one blueprint — a public web app plus two private agents — and a React Native (Expo) app shares the same design tokens for a pixel-identical mobile build.

Challenges we ran into

  • Multimodal at scale. Long videos blew past inline size limits, so we built resumable uploads to the Gemini Files API with poll-until-ready and memory guards — real creator clips, not toy demos.
  • Making one coach feel personal. We built a personalization layer that resolves a persona and playbook per user and injects live context, plus an answer-quality gate so onboarding can't be gamed with gibberish.
  • Real analytics without building N platform integrations — solved with Zernio and a signed-webhook-only ingestion path.
  • Trustworthy coaching. We added deterministic guardrail test suites in CI so the coach diagnoses before it answers instead of hallucinating advice.

What we learned

Right-model-for-the-job beats one-model-for-everything — pairing OpenAI's reasoning with Gemini's vision gave us both sharper strategy and grounded visual feedback. And voice turned onboarding from a chore into a conversation, which is what made the coaching feel human.

What's next

General-availability native launch, scheduled analytics refresh, more platforms, and deeper closed-loop coaching — every post you ship teaches the coach what works for you.

Built With

Share this project:

Updates

Submission history