Book a Call
Client Engagement AI / Developer Tooling

AI Agent Simulation Platform for Iso AI

How Tech Stack Playbook productized Jupyter notebooks into containerized, API-driven simulation environments on AWS — with 12x faster performance and 130 organic SDK installs in 17 hours.

12×
Faster Execution
130
Installs in 17hrs
3
Simulation Environments
Tiered
Product Architecture
Executive Summary

Overview

Iso AI, founded by Jazmia Henry (former AI Engineering Lead at Microsoft), was building simulation environments for unit testing individual sub-components of agentic AI workflows. The technology worked — but it lived entirely in Jupyter notebooks with no API, no web interface, and no product.

Tech Stack Playbook delivered a containerized backend on ECS/Fargate, a serverless API with async task execution, a React web application, Cognito authentication, DynamoDB persistence, and a production architecture that reduced QA execution from 1+ minutes to 5 seconds.

The Challenge

From Research to Product

The AI agent ecosystem had orchestration and observability tools, but nothing that performed component-level testing on individual sub-agents. Iso AI's technology was differentiated — the problem was that it wasn't a product yet.

  • All simulations ran as notebooks or CLI scripts — no containerized runtime or API
  • End users couldn't configure tests without deep technical knowledge and local Python environments
  • No async execution pattern for simulations ranging from 5 seconds to 17+ minutes
  • No scalable compute for 1,000+ parallel iterations needed for statistical validity
  • Python dependency tree exceeded Lambda's 250 MB limit, requiring containerization
  • Without component testing, agents could loop indefinitely generating thousands in API costs
What We Delivered

Production Platform Architecture

The fundamental transformation was conceptual: Jupyter notebooks are for exploration; production systems are for reliability. TSP bridged that gap.

01
ECS/Fargate Containers
Three simulation environments containerized and deployed behind an ALB — 2-3x faster than Lambda for heavy workloads.
02
Async API with Task Polling
POST to submit, GET to poll — handling 5-second to 50-second runtimes with immediate task ID return and status tracking.
03
React Web Application
3-step wizard: select simulation → define data → choose AI model & submit. Run history persisted in DynamoDB sidebar.
04
Authentication & API Keys
Cognito registration with auto-generated API keys, plan tier display, and feature-gated access across Free/Pro/Enterprise.
05
Multi-Model Support
OpenAI, Anthropic, Hugging Face, and Custom model providers — selectable per simulation run.
06
Codebase Containerization
Resolved inconsistencies between notebook and CLI implementations, dependency conflicts, and packaging into repeatable CI/CD.
Key Insight
When Jazmia uploaded the IsoZero SDK to PyPI for a demo video — without sharing the link publicly — it received 130 installs in the first 17 hours, organically validating market demand.
Results

Outcomes & Business Impact

12× Faster Execution IsoZero QA from 1+ minutes in a notebook to 5 seconds on ECS/Fargate.
130 Organic Installs SDK received 130 PyPI installs in 17 hours without public promotion.
Production Platform Containerized backend, async API, web app, auth, and run history — research code → sellable product.
Tiered Product Architecture Free, Pro, and Enterprise tiers with feature-gated simulation access and adversarial testing.
Technical Stack

Technologies Used

Amazon ECS AWS Fargate Flask React Amazon Cognito DynamoDB CloudWatch ECR AWS Amplify CloudFront ALB Docker