Autonomous LinkedIn Content Agent
An autonomous AI agent that generates, schedules, and publishes LinkedIn posts three times a week — pulling live data from a crypto trading dashboard and portfolio CMS, with zero human involvement.
SCREENSHOTS
Research Question
Can an AI agent autonomously manage a LinkedIn content strategy — sourcing real data, generating contextually relevant posts, and publishing on schedule — without any human intervention?
Video Walkthrough
Key Results
- 3 posts per week published autonomously (Mon/Wed/Fri)
- 4 microservices deployed on Railway
- Zero human writing or scheduling involvement
- Post length enforced at 1,300 chars max in code
- Hashtags capped at 5, always at end — enforced in code
- YouTube + GitHub + portfolio links force-injected per tier
- LinkedIn OAuth token valid for 60 days
- Groq LLaMA generation cost ~$0.004 per post
- Tier-specific NOVA signature on every post
Architecture
Case Study
The Problem
Maintaining a consistent LinkedIn presence requires significant time and effort — researching topics, writing posts, scheduling content, and publishing consistently. For developers building in public, this competes directly with actual building time.
The Solution
NOVA is a 4-microservice autonomous LinkedIn agent that handles the entire content pipeline without human involvement:
- Admin service — web dashboard for reviewing and approving generated posts
- Generator service — pulls live data from the crypto dashboard and portfolio CMS, uses Groq's LLaMA model to write tier-specific posts
- Scheduler service — manages the posting calendar (Mon/Wed/Fri) and triggers generation
- Publisher service — authenticates with LinkedIn API and publishes approved posts on schedule
Content Strategy
Three post tiers published weekly:
- Tier 1 (Monday 8AM EST) — Weekly crypto experiment recap, sourced from live trading dashboard data and AI weekly review
- Tier 2 (Wednesday 9AM EST) — Portfolio project spotlight, sourced from portfolio CMS with YouTube build link
- Tier 3 (Friday 9AM EST) — Builder story paired with that week's Tier 2 project, sourced from key findings in the CMS
How NOVA Works
1. Scheduler triggers generation at the right time
2. Generator fetches live data from the appropriate source
3. Groq's LLaMA 3.3-70b writes the post content
4. Post rules are enforced in code — 1,300 char limit, max 5 hashtags at end, links force-injected
5. Post saved to queue with NOVA's tier-specific signature
6. Admin reviews and approves (or sets to Autonomous mode)
7. Publisher posts to LinkedIn via API at scheduled time
Key Findings & Learnings
- Prompt-only rules are unreliable — Groq ignores word limits and link instructions; code-level enforcement is mandatory
- Force-injecting links after generation (not via prompt) is the only reliable way to guarantee they appear
- Tier-specific content pairing (Tier 3 mirrors Tier 2 project) creates a natural Wed/Fri content pair for the same audience
- Weekly AI review endpoint from the crypto dashboard provides richer multi-experiment data than per-decision filtering
- LinkedIn API OAuth tokens last 60 days — refresh logic needed before May 17, 2026
- Content angle rotation must use day_number not internal DB id — insertion order != day order
- Agent personality (NOVA) and first-person voice significantly improves post authenticity over generic "AI posted this" footers