Post Pilot
Team consisting of iOS/visionOS lead Moustoifa Moumini (Voodoo; Swift/SwiftUI, ARKit/visionOS) and Product Owner Alexandre Guirgues (Ship Faster Social; DevOps, C, PHP, SQL).
YouTube Video
Project Description
PostPilot — AI Social Posting Copilot (LinkedIn-first, multi-network ready)
Problem & Target
Founders, PM/POs, and growth leaders know they must post consistently to build authority, but they lack time and workflow. Output is irregular, brand voice drifts, and ROI is unclear.
What it does
PostPilot turns a lightweight client profile (brand, ICP, pains, desired outcomes, voice, CTA) into ready-to-publish LinkedIn posts (text + image) with a pragmatic, secure pipeline:
- Feed: capture product + ICP inputs.
- Configure (AI-assisted): get suggested cadence, topics, hashtags, KPIs, and timing.
- Create: generate preview (strict format and length), attach a relevant Unsplash image, and publish via Zapier → LinkedIn.
Demo flow (60–90s)
• Save profile → get AI suggestions → generate preview (text+image+promo URL) → one-click send to Zapier → LinkedIn post created. Logs and guards validate length (<2,800 chars), place the link before hashtags, and filter sensitive topics.
Execution & Functionality
Production-style FastAPI backend with health checks, structured logging, deterministic guards (length/hashtags/link placement), content filtering, Unsplash fallback queries, and retry/skip mechanics. Works today with a live webhook to LinkedIn via Zapier.
AI & Agents Usage
• GPT-4o: writes native LinkedIn posts and proposes configuration from client inputs.
• Lightweight agent logic: validate constraints, filter sensitive content, assemble final post (insert promo URL before hashtags), and choose a safe image query from hashtags/keywords with fallbacks.
Scalability (One-Person Model)
Minimal ops: AWS EC2 + ngrok + Zapier; SQLite now, swappable to Postgres/RDS. Cron or simple queue can scale scheduling. Easily containerized or moved to serverless later.
Problem Clarity & Market Impact
Solves “no time / inconsistent posting / weak reach” for solo founders, lean agencies, and B2B teams. Delivers consistent voice, predictable cadence, and trackable traffic via promo links (UTMs).
Demo & Product Narrative
From client inputs → AI configuration → compliant post → live publish. Clear, auditable steps; instantly demonstrates value.
Partner Tech Usage
• OpenAI (GPT-4o) for generation + suggestions
• Zapier Webhooks → LinkedIn (Create Update with image + preview URL)
• Unsplash API for license-friendly images (robust query + fallback)
• AWS EC2 (Ubuntu) hosting; ngrok for secure demo tunneling
Technologies used
Python 3.12, FastAPI, Uvicorn, Requests; OpenAI GPT-4o; Zapier Webhooks; Unsplash API; SQLite; AWS EC2; cron; structured logs.
Agent logic, failure modes & safeguards
• Guards: length limits, link-before-hashtags, hashtag detection, blocked-keyword filter.
• Failures handled: LLM overlength → reject & retry; Unsplash 404 → alternate queries; Zapier 4xx/5xx → retry + log; rate limits → backoff + queue.
Next steps
SwiftUI iOS front (Feed/Configure/Create), per-post KPIs, multi-network variants (X/IG/FB/Threads), optional short-video generation (script→voice→captions), JWT auth and roles, Redis queue for scheduling.