Stephen DiCapua

A Relationship Operating System — multi-agent orchestration over a single shared brain

The core product decision: agents shouldn't own their knowledge. Every agent on the platform reads from and writes to one centralized brain, so context captured during acquisition is still there when a support conversation happens a year later. LangGraph handles orchestration; the memory architecture handles continuity.

Orchestration

Six specialized agents — Customer Service, Executive Assistant, Lead Gen, Social Media, Business Analyst, Inventory — built as LangGraph graphs with explicit state, not prompt chains. A Marketing Suite (LinkedIn outreach, SEO, Reddit) runs coordinated campaigns on the same substrate.

Memory architecture

Three vertical layers — Acquisition, Operational, and Relationship Memory — with structured hand-offs between agents. When a lead converts, the lead-gen agent's context transfers to operations instead of being re-learned from scratch.

Platform engineering

Multi-tenant from the first schema: tenant-scoped retrieval, per-business agent configuration, and a deployment topology split across Vercel (product surface) and Railway (agent runtime).

Not a concept deck — the platform is deployed and being validated on Trackply with real users before any client onboarding. Build, ship, measure, then sell.

• LangGraph
• Python
• FastAPI
• Next.js
• Supabase
• pgvector
• Claude API
• Railway
• Vercel

AI-native job search platform — and the production proving ground for Agentic Labs

Built from personal necessity, run as a real product: live users, live agents, live feedback loops. Trackply is where platform hypotheses get tested against reality — which agent behaviors users trust, where hand-offs break, what retrieval quality is actually good enough.

Product surface

Kanban application pipeline, Notion-style workspace, Chrome extension for one-click job capture, smart application builder, and a scam detector for suspicious listings — a complete workflow, not a wrapper around a chat box.

Job Coach Kemba

A LangGraph coaching agent grounded in each user's own documents and history through the shared brain — the same retrieval layer Agentic Labs runs on, exercised by a second product.

Developer experience

MCP-compatible by design: users and developers can pull Trackply context into Claude, ChatGPT, or any MCP-aware assistant. The product participates in the agent ecosystem instead of walling itself off.

Agentic Labs marketing agents run here in production — real campaigns with measurable output, closing the loop between platform engineering and product evidence.

• React
• PWA
• LangGraph
• Claude API
• Supabase
• MCP
• Chrome Extension
• Apify