Heurist AI: Human-in-the-Loop Content Autopilot Agents

Short answer: I built an automated social content agent for Heurist AI that ingested technical updates, summarized them using Claude API, and routed draft posts to a CRM alert channel with interactive 1-click approval buttons before publishing to X and LinkedIn.

Context

Heurist AI is a decentralized AI cloud platform. The primary challenge was maintaining consistent, authoritative technical engagement across social media channels without hiring a dedicated editorial team or risking brand embarrassment through unvetted AI output.

The system

graph TD
    A[Tech Updates & GitHub Commits] --> B[Claude API Summarizer]
    B --> C[State Machine: Draft -> Format]
    C --> D[CRM alert Interactive Approval Gate]
    D -->|Approved| E[Publish to X & LinkedIn]
    D -->|Rejected| F[Feedback Loop]

How I built it

1. State Machine Logic: Structured workflows inside n8n with deterministic states (DRAFTED, FORMATTED, AWAITING_REVIEW, APPROVED, DISPATCHED).
2. Error Recovery & Circuit Breakers: Built retry loops with exponential backoff for LLM rate limits and API timeouts.
3. Interactive Approval Gate: Dispatched formatted previews with 1-click Approve/Edit/Reject buttons into a dedicated CRM alert channel.
4. Distribution Automation: Routed approved payloads to X API and LinkedIn marketing endpoints with native media attachments.

Results

• Maintained regular top-of-funnel technical presence across X and LinkedIn without a dedicated content department.
• Designed agent workflows with resilient state-machine logic, retries, and comprehensive error handling.
• Ensured 100% of outbound social posts passed through human editorial verification before publishing.

See the underlying workflow: Waterfall Enrichment

Hire me for this motion