Skip to content
back to projects

~/projects/tangle-ai-platform

AI Platform

Tangle AI Platform

How I worked across four AI applications and the shared systems for workflows, retrieval, isolated development environments, capacity, and production delivery.

TypeScriptAIWorkflowsRAGDockerNix
project.json
Organization
Tangle
Period
Mar 2024 — Sep 2026
My role
Core developer across Blueprint Agent, Sandbox, Tangle ID, and Tangle Router; builder of shared product and platform systems.
Blueprint Agent, one of the products built on the Tangle AI platform

01 // the platform problem

The platform problem

The product suite combined user-facing AI applications with shared infrastructure for workflows, retrieval, identity, isolated execution, model access, and production operations. The challenge was not a single feature: each product needed dependable platform behavior without every application rebuilding the same systems.

Sandbox environments also needed to arrive as complete AI-development workspaces with tools and user data, rather than as empty containers that required additional setup.

02 // my product and workflow work

My product and workflow work

  • Built web and backend features across Blueprint Agent, Sandbox, Tangle ID, and Tangle Router.
  • Built shared libraries used across the product suite and led workflow development across the AI applications.
  • Built production RAG for Blueprint Agent and Sandbox, using indexed retrieval for on-demand context and embedding-backed storage and retrieval in MCP servers.

03 // sandbox infrastructure

Sandbox infrastructure

  • Built the runtime that launched complete development environments with preconfigured tools and user data.
  • Built automatic capacity scaling that deployed additional backend servers before existing sandbox and container capacity was exhausted.
  • Worked across provisioning, sessions, WebSocket coordination, authentication, and failure recovery rather than treating the sandbox as a bare-container problem.

04 // production delivery

Production delivery

  • Built deployment flows that verified replacement services before switching production traffic.
  • Supported rollback from deployment snapshots and added session recovery and failure-path regression coverage around shared infrastructure.
  • Kept workload and user-count claims out of this case study where the available figures mix internal activity with external use.