Skip to content

SupplyScribe
CLI agent that automates operational marketing — faster posts, less manual work.

CLIENTSkład Muzyczny
TERM2024-05
ROLELead Developer / Architect
STATUSCompleted

"SupplyScribe turned hours of writing posts into minutes of approvals."

SupplyScribe: Automating content marketing with LLMs and LangChain

SupplyScribe is a simple CLI script (AI agent) that automates the operational marketing workflow at Skład Muzyczny. It aggregates delivery and event data, then generates tailored posts for the blog and five social platforms.

📋 Project metrics

  • Start: May 12, 2024 (Initial Repo)
  • Go-live: June 2024
  • Status: Completed / Production (Skład Muzyczny)
  • Role: Lead Developer / Architect
  • Goal: Shorten time-to-distribute updates about deliveries and events across all channels

🚀 Project evolution (AI workflow)

  1. Aggregation (Ingestion) — connect to Retable API (gear deliveries) and the events calendar.
  2. LLM orchestration (Processing) — LangChain as the central interface; models from OpenAI, Anthropic, Groq.
  3. Formatting (Prompting) — dedicated prompts for FB/IG/LinkedIn/X/Google Business tuned for tone and length.
  4. Delivery — drafts saved to AITable and optionally pushed to Ocoya for publishing.

🎯 Business problem

Classic bottleneck: a new delivery required writing a blog post plus five social variants. More channels meant linearly more time.

❌ Pain points

  • Delays (Cost of Delay): “Delivery arrived, but no post yet” because there’s no time to write.
  • Scattered sources: Deliveries in Retable, events in the calendar — constant context switching.
  • Inconsistent formats: Six platforms → uneven quality and writer fatigue.
  • Creativity drain: Hours rewriting specs instead of focusing on strategy and campaigns.

💡 Why it works (product approach)

  • Assistant, not a replacement: LLM drafts are based on hard data from Retable; publication always after human approval (Ocoya/AITable) → keeps hallucinations in check.
  • Multi-model strategy: LangChain makes it easy to test models (OpenAI, Anthropic/Claude, Groq) and balance quality vs. cost.
  • Prompt separation: Files in src/prompts/ per channel for easy tone tweaks without affecting others.

📈 Impact (ROI)

Automation shifted the workflow from “I write” to “I approve.”

MetricEffectBusiness value
Time savings~3 hours weeklyMore time for strategic work
Time-to-marketDrafts appear immediately after deliveryFaster info → faster sales
Communication regularityNo communication gapsBetter reach in social algorithms

"SupplyScribe turned hours of writing posts into minutes of approvals. The script doesn’t replace me, but it massively helps daily work."

🛠️ Architecture & tech stack

  • Core: Node.js + TypeScript (modular architecture with a central coordinator)
  • AI & orchestration: LangChain (@langchain/core, @langchain/openai, @langchain/anthropic, @langchain/groq)
  • Ingestion: Axios, Retable API (deliveries), Calendar API (events)
  • Distribution: APITable (AITable) + Ocoya API (content distribution)

🚀 Summary

A pragmatic AI agent connecting operational and marketing systems. Shows how to build a GenAI agent in Node.js to solve a concrete business problem. Essentially, this is a sketch of a tool that can later be used in other projects, but it's very helpful in daily work.

Contact

Have a similar challenge? Write to me — I will come back with a proposal for next steps.

Send message