Case Study
Built the core engineering infrastructure for an AI agent platform that routes natural-language goals to domain-specific execution modules — not just conversation, but real outcomes.
Overview
Relaystack is an AI platform that executes real-world outcomes through composable, installable skills — not just conversation. I joined as a full-stack engineer responsible for building the core agent infrastructure: the system that actually makes things happen when a user gives the AI a goal.
The constraint
Most AI products stop at generating text. The hard part is building agents that reliably dispatch tasks to the right modules, maintain context across sessions, and execute multi-step workflows without losing state or producing unpredictable output. Every skill needs to be testable, recoverable, and auditable.
Highlights
Engineered a TypeScript/Node.js system that parses natural-language goals and dispatches tasks to domain-specific AI modules. Each skill exposes a typed contract so the router can match intent to capability without ambiguity — eliminating manual prompt loops entirely.
Implemented a two-tier memory layer: Redis for fast session state (what the agent is doing right now) and PostgreSQL for long-term context (what it has learned about a user's goals over time). This enables genuinely personalised, context-aware responses across sessions rather than starting from scratch each time.
Integrated OpenAI tool calling to power multi-step execution pipelines — the agent can draft content, identify prospects, and schedule follow-ups all within a single session, with each step informed by the results of the last.
Established a TypeScript monorepo with shared Zod validation schemas across frontend and API layers. This single source of truth for data contracts reduced interface-mismatch bugs to near zero and significantly accelerated feature delivery for the broader team.
Stack
TypeScript · Node.js · React · Redis · PostgreSQL · OpenAI API · Zod
Relaystack is currently building toward its first production release. The infrastructure is designed to support an ecosystem of installable skills — each one a composable unit of real-world capability.