Back to Work
6-month rebuild · 30 microservices
Scaling an AI-Native Engineering Org
Led 11–13 engineers across Frontend, Backend, QA, and AI/Prompt Engineering through a 6-month rebuild of SalesAI's core platform. Paired Shape Up cycles and DORA metrics (MTTR, Change Failure Rate) with automated Playwright/Jest/JUnit CI/CD and LaunchDarkly feature flags.
Engineering Leadership · Shape Up · DORA Metrics · AI Teams
The Problem
SalesAI needed a full-stack rebuild of its core AI sales platform. The rebuild had to happen on a fixed timeline, while the team kept shipping real-time AI features to customers.
Architecture & Approach
- Ran delivery on Shape Up: fixed-time, variable-scope cycles with shaped pitches and hill charts, so scope tradeoffs were explicit rather than discovered late.
- Architected an event-driven system on GCP: 30 microservices, Cloud Pub/Sub async processing, MySQL/Redis consistency patterns, and Terraform/Terragrunt across 42 modules.
- Built velocity infrastructure: an NX monorepo with 40+ projects, auto-generated TypeScript API clients, and 18 CI/CD pipelines running Playwright, Jest, and JUnit suites.
- Shipped behind LaunchDarkly feature flags and tracked DORA metrics (MTTR, Change Failure Rate) as team-level signals, not individual scorecards.
- Integrated multi-modal AI: OpenAI GPT-Live real-time voice, LiveKit agents, conversational SMS, and automated data extraction, alongside HubSpot and Twilio.
Measurable Impact
- Delivered the full-stack rebuild in 6 months: a React 19 / Next.js 16 frontend backed by 30 microservices.
- Enforced a 70% test coverage bar across the monorepo.
- Expanded the platform from a single channel to multi-modal AI engagement (voice, SMS, data extraction).
- Shipped a low-code visual workflow builder with 9 step types and 7-way call outcome routing for non-technical users.