Reference systems, not slideware.
Internal reference systems and prototypes that show how we build production AI — architecture, what gets logged, where evals run, and the failure modes we design for. Clearly labelled; not client case studies.
AI Ops Dashboard
A control layer for agent runs, costs, evals, approvals, and failures.
RAG Pipeline
Ingestion to hybrid retrieval to reranking to cited answers.
Production Agent
A bounded LangGraph agent with explicit state, tools, and approvals.
Claude Code Workflow
Repo instructions, skills, MCP, and a GitHub Actions review loop.
MCP Server
A TypeScript MCP server exposing internal tools behind auth.
Model Routing System
Routes tasks across OpenAI, Claude, Gemini, and Grok with fallback and cost control.
DVNC Experiments
Working interfaces for operating decisions
Standalone concepts where interaction is part of the argument. Each experiment is public, clearly bounded, and separate from client work.
Public concept · Version 3
Authority Relay
Choose whether one fictional refund request should be answered, actioned, approved, handed to a person, or refused. The interface shows why those outcomes need different authority and evidence.
Open Authority Relay- Scenario
- Fictional
- Production data
- None
- Lead capture
- None
- Action backend
- Not connected
Public concept · Version 1
Context Strata
Rewind one synthetic maintenance-window memory as a source is observed, corrected, expired, and contradicted. The interface withholds retrieval when the current ticket and runbook diverge.
Open Context StrataInspectable states
- 01Observed
- 02Superseded
- 03Expired
- 04Divergent
Synthetic scenario. No production data, model call, storage, lead capture, or action backend.
One letter, every week. Working systems — not hot takes.
Build logs, agentic engineering decisions, agent failures, evals, and what survives real users. Sent weekly, never more.