Back to Work
~150k lines of TypeScript, built solo
MeetingSync
Architected a multi-tenant AI SaaS platform (~150k lines of TypeScript, 1,000+ commits) for engineering 1:1s and team alignment, with provider-agnostic LLM routing and MCP integration.
Node.js · PostgreSQL · Redis · Multi-Tenant SaaS
The Problem
1:1s, team syncs, and OKRs usually live in separate tools. Context and action items get lost between them, and AI features built on a single model provider are expensive and fragile.
Architecture & Approach
- Built multi-model LLM routing (OpenAI, Anthropic, Gemini) by task, tier, and token cost, with response caching and deterministic non-AI fallbacks.
- Shipped natural-language scheduling, Zoom transcript action-item extraction with confidence scores, and an in-app agentic assistant.
- Integrated bidirectionally with Google Calendar, Microsoft 365 Graph, and Slack, and added MCP integration for AI tooling.
- Hardened tenant isolation for SOC 2 readiness with field-level encryption, MFA, and self-run security audits.
- Defined Stripe tiered pricing, meeting-quality scoring models, and a GTM/beta acquisition plan.
Measurable Impact
- Built solo end to end: ~150k lines of TypeScript across 1,000+ commits.
- Shipped to production as a multi-tenant SaaS with tiered billing.