MediaRaq — Engineering¶
MediaRaq is a multi-tenant agentic AI platform in production. AI personas run as tool-using agents. They hold conversations, remember people and business knowledge through retrieval-augmented memory, capture leads and propose appointments for human approval, and produce and publish video, images and articles. It was designed, built and operated by Michael Inman, directing a fleet of AI coding agents under written, machine-checked rules.
This site publishes MediaRaq's engineering documentation. The source code is private. Product: https://mediaraq.com
Start here¶
For hiring managers and engineering leaders¶
- What's in production and what's in pilot: Agentic AI capabilities and their status
- How the agents work: Persona agent architecture, Memory and retrieval, Model routing
- A real evaluation result: Evaluation. The retrieval release gate failed, the launch was held, the fix was verified 10/10 in production.
- How I build with AI, and where my judgment changed the result: AI-assisted engineering and human oversight
For investors and acquirers¶
- What the platform does: What MediaRaq is and the system context
- Maturity and risk: Validation and risk control, Limitations and trade-offs
- Where it scales and what it costs to grow: Deployment topology
- Where it's going: Roadmap
The documents¶
| Document | What it covers |
|---|---|
| Engineering Summary | Capabilities and status, decomposition, AI-assisted process, evaluation, risk, scenarios, trade-offs |
| Architecture Addendum | Components, agent and memory architecture, approvals, MCP, routing, pipelines, security, deployment, decision log |
| Roadmap | Shipped, in flight, next, later |
In numbers¶
18 backend services · 14 shared libraries · about 6,000 automated tests · 25 event topics · 69 MCP tools · 4 LLM providers · about 740 gated handoffs and 680 validation records since April 2026