Daniel Estevez
AI Systems Engineer
- LOCATION
- Thailand · Remote
- CV
- Daniel_Estevez_CV.pdf
- LINKS
- STACK
- Agent Harness DesignMulti-Agent OrchestrationModel Context ProtocolLangGraphMulti-Vendor LLM RoutingClaude APIVercel AI SDKLLM EvalsStreaming LLMsExecution State MachinesTypeScriptReactNode.jsPostgresOpenAI APIAI-Augmented DevelopmentGraphQLEnvelope Encryption (BYOK)System ArchitectureSpec-Driven Development
Software engineer with 15+ years of experience. I build systems that scale and I direct AI agents to build them — I never ship code I don't fully understand.
Background
15+ years shipping production software on backend foundations — SaaS platforms, real-time systems, cross-platform applications, Web3 — owning architectural decisions end to end. At Game7 from November 2023 to March 2026 I designed multi-agent AI content generation with intelligent model routing, built a custom evaluation framework for non-deterministic outputs, and established the team's AI-augmented development workflow. Independent since 2026. Four years working with AI every day.
How I Work
I design systems where agents and humans build side by side. Structured specs as implementation briefs, mandatory non-author review of agent-written code, strict tech-debt and context discipline. Every line shipped is fully understood by the human who owns it. Fifteen years of engineering judgment is what makes the AI part work — I direct agents rather than defer to them.
Vinaya
Four years of building with AI agents taught me the hard part is never the models. It is keeping a codebase honest while something else writes most of it. Vinaya is the harness I built for that: one path for every change, deterministic checks that give an agent the same verdict on its own machine and in CI, and a state model where status is never stored — only derived from what actually happened, so work that is not really done cannot read as done.
Atta Labs
My lab — where all of this actually runs. My own agent teams, written as specifications and compiled into executable flows. My own multi-LLM engine, routing across Claude, Gemini, GPT and Grok with cost, quality and fallback logic. My own Model Context Protocol (MCP) server with bearer auth and envelope-encrypted keys. Vinaya governs how every line of it gets built, and this page is one of the products running on it.
It's all open — read the code, and read the issues.