LinkedInIA · interactive architecture

The LinkedInIA architecture, from A to Z

Automated LinkedIn posts and job applications. The AI does everything; the human only validates, edits or rejects.

Hover over any module of the diagram: a detailed explanation appears
0
interview appointments
obtained in 15 days thanks to automated resume sending
0
resumes sent
automated applications (resume + letter)
0
profile posts
published, with infographics
0
feed posts
comments published on the feed
0
feed replies
replies to received reactions
0
cumulative interactions
on joined conversations (feed)

Real platform counters, as of July 3, 2026.

«

25 interviews. Zero job offer.

No offer, because I "can't code". Recruiters look for people who know how to use AI, then ask that same person to do the AI's work. The human is a thinker, an architect: no one can compete with an AI that writes 80,000 lines of clean code in under an hour. It is like demanding that a Formula 1 driver know how to build the engine. The driver wins the race: he picks the line, manages the risk, decides. The engine is built elsewhere. LinkedInIA was born from this insight: the AI does, the human thinks and decides.

Laurent Poupet · OpenCenterAI architect
LinkedInIA · Technical Architecture Automated LinkedIn posts + job applications · human + AI pair · 100% human validation CAMPAIGN TYPES Feed posts (comments) Profile posts (+ infographic) Recruiting (resume + Easy Apply) Events (invitations) LinkedIn (the network) LEGEND zone: client machine zone: OVH server AI operations human control API · protocol internal link DB access (read / write) Client / Owner validates · edits · rejects HUMAN CONTROL POINT CLIENT MACHINE Chrome browser + Chrome MCP connector drives the client's real LinkedIn session scan · publishing · Easy Apply · invitations AI Pair · Claude Claude Code session · the operator · runs the skills received from the MCP server Claude Code Claude Design creates the profile-post infographics (claude.ai/design) Campaign auto-config the client says what they want (topic · goals · plain language) if in doubt: multiple-choice questions generates the tailored campaign MCQ questions Web cockpit + iOS app the client validates, edits or rejects statuses: pending → validated → published + superadmin backoffice (multi-tenant) OVH SERVER MCP server (mcp-opencenterai) the pair's tools + live instructions TypeScript · Node.js Resume generation tailored resume + cover letter screening questions stored in the database (per job offer) Skills .md the pair's skillset · linkedinia-binome · post-image-publishing · job-application · event-campaign Application server (ia-academy) renders the resume PDF (/li/cv) generates the infographic image OG pages for tracking links (/li/code) serves the web cockpit + admin API Next.js · TypeScript MongoDB campaigns · client configs drafts: posts · comments resumes + cover letters screening Q&A bank audit log (traceability) pair ↔ dev liaison 100% managed by the AI SSH TUNNEL · encrypted link client ↔ server Voyager API (LinkedIn internal API) 1 · Scan   4 · Publish MCP protocol (tool calls) · 2 · Drafting authentication: unique key (one per client) instills the skills (.md) into the AI pair 4 · Publish (drives) REST API (HTTPS) 3 · Validation validates / edits / rejects + states their needs (plain language) reads / writes · 5 · Tracking (audit log) renders the PDF internal HTTP API creates the infographic A CONTENT'S JOURNEY · FROM A TO Z The AI does everything · the human only: validates · edits · rejects 1 · SCAN (AI) the AI pair reads the LinkedIn feed via the Voyager API 2 · DRAFTING (AI) anti-template draft + signature stored in DB (status: pending) 3 · VALIDATION (HUMAN) the client validates, edits or rejects web cockpit or iOS app 4 · PUBLISHING (AI) real action in Chrome (Chrome MCP) strict verbatim of the validated text 5 · TRACKING (AI) replies to received reactions audit log in DB

No scraping. Human-like navigation.

LinkedInIA does not scrape pages and does not hijack any server-side API. The AI navigates inside the client's real Chrome browser, through Chrome MCP: it scrolls, reads, clicks and fills forms like a human hand, in the real LinkedIn session. On screen, it is the orange halo cursor: when it moves, the AI is navigating.

client's LinkedIn session · driven by Chrome MCP
Comment
Comment
Comment
Animated demonstration: the Chrome MCP orange halo cursor, the AI's human-like navigation.

The interfaces, for real

Left: the real web console (client access only). Right: the iPhone app: the client validates, edits or rejects from anywhere.

stats.opencenterai.com
The real web console · client access only
To validate: the human gesture
To validate: the human gesture
To publish: publishing queue
To publish: publishing queue
Publications: the history
Publications: the history
Stats: network + engagement
Stats: network + engagement