AI LinkedIn Content Automation Engine

Project Overview

Developed during my intensive 1-month AI internship at Mirai School of Technology. The system automates the time-consuming process of scanning technical news feeds, verifying contextual relevance against past curated themes, and synthesizing publication-ready professional drafts for review.

Problem Statement & Context

Business Objective

Create a proof-of-concept autonomous research and drafting agent that compresses the content generation lifecycle from hours to minutes while preserving brand voice and ensuring zero factual hallucination through grounding in verified real-time sources.

Methodology & Architecture

1. Trigger-based scheduling via n8n cron node. 2. Query formulation and real-time tech article retrieval using Tavily Search API. 3. Semantic vector search in Pinecone to compare against established writing tone and past published topic clusters. 4. Few-shot prompt chaining to generate compelling hooks, body insights, and actionable conclusions. 5. Human-in-the-loop dispatch to Discord/Slack for one-click approval before posting.

Approach & Implementation

The workflow operates on a visual n8n canvas. Inbound topic requests trigger an asynchronous execution tree: Tavily retrieves clean markdown extracts of high-authority technical sources; Pinecone computes cosine similarity with previous high-performing posts; and a structured JSON schema guarantees consistent output formats.

Data Sources & Ingestion

• Tavily Search API real-time web crawler
• Pinecone indexed brand-voice embeddings

Key Findings & Observations

Business Recommendations

• Always incorporate a human review gate rather than fully unassisted automated publishing to preserve personal brand authenticity.
• Maintain a localized vector index of audience interaction metrics to train future prompt templates on high-resonance topics.

Measurable Results & Deliverables