AI / ML · SaaS
Tryneth
Tryneth is a multi-tenant AI marketing and CRM platform: a customer connects their website and gets a team of AI agents on top of a full CRM, lead capture, session recording, and multi-platform publishing. I own large parts of the backend, from AI-agent orchestration and usage-based billing to the lead pipeline and production hardening.
- Role
- Backend & AI engineering
- Client
- Techneth
- Timeline
- Ongoing
- Year
- 2026
AI-agent orchestration and usage-based billing for 5 purpose-built agents
Telemetry-driven lead capture with bot filtering and resilient inserts
Row-Level Security across 97 of 98 production tables
Overview
Tryneth connects a company's website to an AI team (a blog writer, a voice assistant, and analysts) sitting on top of a real CRM. My work is the engine room: a large TypeScript backend, around 56,000 lines across roughly 180 API endpoints and 13 scheduled jobs, powering AI orchestration, billing, the lead and telemetry pipeline, integrations, and reliability.
The challenge
A multi-tenant SaaS that charges by usage has to be exact. Every AI action needs to be metered without ever double-charging, leads have to be captured even when tracking scripts send messy data, and one customer's data must never leak into another's.
On top of that, AI calls are expensive, so cost control and caching matter as much as correctness.
What I built
I built the orchestration layer for five AI agents with atomic, usage-based metering, using a claim-first compare-and-swap approach so a request is never billed twice. I rebuilt lead capture around telemetry with bot filtering and schema-resilient inserts so no lead is lost, and consolidated content publishing behind one orchestrator across seven publishers, including five CMS integrations.
I also cut a major cost centre by fixing prompt caching on the busiest endpoint, eliminating roughly ten million uncached tokens, and hardened production: Row-Level Security on 97 of 98 tables, health probes, graceful shutdown, and closing seven critical security findings.
Outcome
Tryneth runs as a reliable, multi-tenant platform where AI features are metered accurately, leads are captured cleanly, and the whole system stays fast and secure under real traffic.
Key features
An orchestrator that runs five purpose-built AI agents behind one interface
Usage-based billing with atomic, double-charge-proof metering
Telemetry lead capture that survives messy tracking data and filters bots
Content published across seven destinations, including five CMS platforms
Prompt caching that removed roughly ten million uncached tokens on the busiest endpoint
Production hardening: Row-Level Security, health probes, and graceful shutdown
Built with
Gallery
Have a project like this?
I take products from first sketch to shipped, design, build, and the AI in between. Tell me what you’re working on.
