Linspire: LinkedIn Content Engine
An AI content platform that turns a LinkedIn profile or website URL into a full publishing workflow: brand analysis, voice-matched post generation, scheduling, and engagement pods, built as an own product from scratch.
The Challenge
Professionals know LinkedIn drives business, but consistent, authentic content is hard. Most people either stare at a blank page or lean on generic AI tools that produce posts which sound like everyone else: generic, repetitive, disconnected from their actual brand, their audience, and what's happening in their industry right now.
What Was Built
Linspire is an AI-powered LinkedIn content platform that turns a single input (a LinkedIn profile PDF or a company website URL) into a full content engine. It analyzes the user's brand and voice, then generates, schedules, and publishes posts that sound like the person who would have written them, grounded in their positioning, audience, and current trends.
Pro and Elite plans add engagement pods: groups that auto-like new posts within a natural 30-60 minute window to organically boost reach.
How It Was Solved
The platform runs on FastAPI + Angular 19, PostgreSQL via raw asyncpg SQL (no ORM), orchestrated by LangGraph pipelines and a multi-provider LLM layer (LiteLLM, OpenAI default with Anthropic fallback).
Brand analysis
Runs through two fixed LangGraph DAGs: a 5-node pipeline for LinkedIn PDF analysis (parse → validate → structure → analyze → suggest improvements) and a 6-node pipeline for website analysis (scrape → validate → identify pages → scrape subpages → optimize → extract insights). Inside those pipelines, three genuine tool-calling agents (ProfileAnalyzer, SiteScout, BrandAnalyst) run bounded reasoning loops, deciding which tool to call and in what order: scoring profile sections against industry benchmarks, or picking which website pages are worth scraping before committing to the crawl.
Content generation
Powered by a Writer Agent running a five-phase lifecycle: it loads only the context it needs (brand analysis, voice profile, ICP, RAG over past posts via pgvector, live trending topics via web search), drafts three posts with a stronger model, evaluates them against quality checks (hook strength, length, voice fit, ICP alignment), revises any that fail, and returns structured output. A two-model routing pattern keeps orchestration cheap while creative writing uses the stronger model. All user-supplied context is sandboxed against prompt injection.
Infrastructure and billing
The system is credit-gated across four subscription tiers (Free/Starter/Pro/Elite) with Stripe billing. Three AWS Lambda workers handle async jobs: scheduled publishing, engagement-pod likes, and monthly credit resets. Everything is traced through Langfuse and deployed on EC2 (backend) and AWS Amplify (frontend).
Results
Linspire delivers a complete, end-to-end pipeline: from a single PDF or URL upload to a published, engagement-boosted LinkedIn post, with no manual drafting required. The architecture cleanly separates fixed orchestration (LangGraph DAGs, predictable and auditable) from genuine agentic reasoning (tool-calling loops where autonomy adds real value: page selection, brand synthesis, content quality checks).
Want to build something like this?
I help founders take ideas like this from zero to a live, AI-ready product.
Book a Call