← oshalusirelaystack.io
Case study

Relaystack

An AI platform that builds and runs workflows with agents. I'm the founding engineer, and I lead the team that shipped Social Growth Manager.

Founding engineer · Jan 2026–present · Relaystack · Web
Overview

Relaystack builds AI agents that carry a job across several systems, and stop for a human when the next step needs a decision.

The first operator is Social Growth Manager. It works on LinkedIn and X, and each account can run in one of three modes: draft only, draft and wait for approval, or full autopilot.

I set the technical direction with a team of four engineers and designers. I also built the orchestration and execution layers the operator runs on.

The constraint

A social post that goes out wrong is public. The same workflow has to support someone who wants a draft, someone who wants to approve every send, and someone who wants it handled. Waiting has to be a real state of the run, not a missing feature.

Highlights

Social Growth Manager

The first operator on the platform. It drafts and publishes for LinkedIn and X. The mode is a setting on the account: draft, approval, or autopilot.

Orchestration that survives a partial failure

A run is a sequence of steps, not one prompt. The engine orders those steps, retries what failed, and picks a run back up when it dies halfway through a multi-system task.

An execution layer that keeps state

Workflows call external tools and APIs. The execution layer holds the context and the state of a run, so the next step knows what the last one already did.

Stack
Next.js · TypeScript · Node.js · Turborepo · BullMQ · Redis · PostgreSQL · OpenAI & Anthropic SDK · Docker · AWS · GitHub Actions

Social Growth Manager is live. The same engine is what the next operators will run on.