Local Core – AI Content Engine

Pipeline Architecture

RSS Feeds + Google News + Hacker News
              ↓
    Content Ingestion (100+ candidates)
              ↓
    Viral Scoring Engine (relevance + tech keywords)
              ↓
    "King" Selection (highest scored article)
              ↓
    LLM Post Assembly (Gemini Pro/Flash)
              ↓
    Visual Generation (Multi-provider cascade)
              ↓
    LinkedIn API Publishing

Project Use Case & Execution

The core of this system is an autonomous Django management command (publish_daily_highlight) designed to run daily via cron. It completely automates the tech news lifecycle—from discovery to social media publication—ensuring a consistent, high-quality social presence without manual intervention.

The Daily Autonomous Pipeline

1. Stateless Purge: Wipes all temporary article candidates from previous runs to ensure fresh processing.
2. Multi-Source Ingestion: Autonomously fetches the latest data from RSS feeds (OpenAI, Cloudflare, Anthropic), general Web Discovery, and the top 50 Hacker News stories. (GitHub and Reddit ingestions are intentionally bypassed to reduce noise).
3. Filtering & Scoring: Evaluates all ingested content from the last 24 hours against strict developer-focused heuristics, filtering out consumer tech noise and assigning a viral relevance score.
4. "The King" Selection: Automatically isolates the single highest-scoring article to become the day's highlight.
5. AI Content Generation: Passes the selected article to an LLM (Gemini) to generate an engaging, professional LinkedIn post, enforcing a strict "Observer" voice.
6. Dynamic Image Generation: Prompts the AI to generate a unique, context-aware image specifically tailored to the article's technical theme.
7. LinkedIn Publication: Connects to the LinkedIn Official API to seamlessly publish the text and image directly to an Organization Page (or personal profile as a fallback).
8. Final Cleanup: Operates as a completely stateless pipeline by wiping all temporary database rows after a successful post, relying only on external URLs for future deduplication.

Safety & Quality Mechanisms

• Execution Locks: Requires a hardcoded safety authorization code (--code=EXECUTE_PUBLISH_NOW) to prevent accidental or unauthorized triggers.
• Fail-safes: Automatically falls back to text-only publishing if image generation fails, ensuring the daily post always goes out.
• Strict Guardrails: Prevents the AI from using corporate "We/Us" pronouns and auto-appends source links to maintain credibility.

Tech Stack