Projects

Sector: Own IP

Production

cortex-ai — multi-agent orchestrator

The engine that serves this site’s chat. Verifiable live, right here.

Try it in the chat
Architecture diagram

cortex-ai is a multi-agent, multi-modal orchestrator in Go that I wrote for my own operation. It is not a wrapper over an SDK: it is an engine with vector memory and AI providers abstracted behind a common interface.

The same brain that serves this portfolio’s chat is the decision core of a real product. It is the hardest-to-copy differentiator on this page: I don’t claim it, you’re using it.

Key metrics

Go, monorepo
~48 KLOC Go, monorepo
Microservices · 69 packages
9 Microservices · 69 packages
Covered by tests
~40% Covered by tests
Several AI providers, abstracted
Multi Several AI providers, abstracted

Stack

  • Go
  • AWS Lambda
  • Function URL
  • pgvector
  • RAG
  • CQRS

Decisions and trade-offs

  1. Serverless Lambdalith architecture

    A Go core serves the whole API from AWS Lambda via a Function URL: no persistent server, low cost, and portable between a VPS and Lambda with no conditional branches. Command/read separation (light CQRS) isolates writes from reads.

  2. AI providers behind one interface

    Several AI providers sit behind a common abstraction: swapping model or provider never touches domain logic. Clean/hexagonal with strict isolation of external SDKs.

  3. A "two-brain" memory

    RAG over embeddings plus episodic memory (pgvector): the agent recalls recent context and retrieves relevant knowledge without dragging the whole history into every call.

  4. One engine, two uses

    cortex-ai is at once a product’s decision core and this site’s chat engine. One asset, reused: the signal that I build platforms, not one-off features.