Sebastián García-Moreno Zinchenko

Sebastián García-Moreno Zinchenko · AI / Agent Engineer

I build production-grade agentic systems.

Multi-agent workflows, grounded RAG, evaluation harnesses, model routing, and observability for LLM systems that need to work beyond demos.

Agentic Works / A field guide

How agency takes shape

01 / Field

Intent precedes intelligence

Before an agent can act, ambiguity must become context, constraints, and direction.

02 / Architecture

A model is not an agent

Agency emerges when reasoning connects to tools, memory, state, and the world it can change.

03 / Threshold

Capability needs a boundary

Permissions, guardrails, and human control define where autonomy may begin—and where it must stop.

04 / Evidence

Trust is built in the loop

Evaluation, observability, and recovery turn plausible behavior into a system that can be relied on.

05 / Agency

Autonomy should return control

The strongest agents act with purpose, coordinate without losing context, and leave every decision inspectable.

Public signals, not adjectives.

Open repositories and merged contributions provide the first layer of evidence.

OSS

MERGED

promptfoo contributor

Merged a CLI rendering fix and validator test-coverage work in the LLM evaluation and red-teaming framework.

RAG

PUBLIC

CodeRAG

Hybrid retrieval, verified code citations, code-graph evidence, and CLI, REST, MCP, and local web surfaces.

HARNESS

PUBLIC

omp-pantheon

A local agentic-engineering harness with specialist agents, SpecSafe discipline, EvalFly evidence gates, durable memory integration, trace hygiene, and branch-level E2E verification.

The systems I can help a team ship.

Engineering strengths framed around the work an AI product team needs: orchestration, grounding, evaluation, and operation.

  1. Agentic workflows

    I build multi-agent pipelines and self-hosted runtimes with tool use, MCP/A2A interfaces, channel adapters, and operational workflows.

    • Multi-agent orchestration
    • MCP and A2A
    • Operational workflows

  2. Grounded AI systems

    I design RAG systems that retrieve source-backed context, verify citations, and expose evidence to both users and agents.

    • Hybrid retrieval
    • Verified citations
    • Source-backed context

  3. Evaluation and reliability

    I use evaluation harnesses, red-teaming, and test-first workflows to make AI-assisted work reviewable against explicit criteria.

    • Evaluation harnesses
    • Red-teaming
    • Test-first delivery

  4. Observability and operations

    I instrument traces, token cost, latency, and errors, and design model-routing and provider-fallback paths for graceful degradation.

    • LLM observability
    • Model routing
    • Provider fallbacks

Three repositories. Three operating problems.

These public repositories show approaches to grounded retrieval, reviewable agentic engineering, and self-hosted agent operations.

Public repository

CodeRAG

Problem
Codebase answers need source references that can be checked instead of plausible but ungrounded citations.
System
A verified context engine that indexes repositories, retrieves hybrid code context, and verifies citations against indexed snapshots.

Evidence

  • Python package and CLI
  • REST and MCP surfaces
  • Verified citations and context packs

Public repository

omp-pantheon

Problem
AI-assisted engineering needs contracts and evidence that make work inspectable before it is called complete.
System
A local OMP-based engineering harness that combines specialist agents, SpecSafe, EvalFly evidence, durable memory, and branch E2E verification.

Evidence

  • TypeScript
  • EvalFly evidence gates and opt-in local enforcement
  • PASS / FAIL / INCONCLUSIVE branch verification

Public repository

AgentForge

Problem
Self-hosted agents need a persistent runtime, operational tooling, and adapters beyond a single chat surface.
System
A TypeScript monorepo for operating self-hosted agents with a persistent Mastra runtime, Convex-backed data, and CLI-driven workflows.

Evidence

  • HTTP, Discord, and Telegram channels
  • MCP and A2A primitives
  • CLI, runtime, core, and dashboard packages

How I make AI work reviewable.

The standard is not a persuasive demo. It is a system whose sources, behavior, and failure paths a team can inspect.

  1. Ground every answer in inspectable source.
  2. Replace confidence with evaluation evidence.
  3. Design failure paths before the happy path ships.
  4. Keep model and tool boundaries explicit.
  5. Use human review where ambiguity is expensive.
  6. Treat agent operation as engineering, not prompting.

Professional trajectory.

Recent roles span agent engineering, software systems, conversational AI, and automation.

  1. Present

    Agentic Engineering Agency

    AI / Agent Engineer

    I operate the technical stack for an early-stage AI engineering studio, leading client builds and internal tooling across agent pipelines, RAG, evaluation, harness engineering, and observability.

  2. Present

    Voblakye

    AI / Agent Engineer

    Ongoing software and AI engineering work across internal initiatives.

  3. TendencIA

    AI & Automation Developer

    Built WhatsApp and Telegram bots, customer-support chatbots, and internal automations using third-party LLM APIs. For AIRE, validated and integrated AI-backed flows and owned most of the Flutter and Supabase frontend.

Let’s build AI systems that hold up beyond the demo.

Open to remote AI / Agent Engineer roles across US and European time zones.

Based in Guadalajara, Mexico, with EST and CET overlap. If you are hiring for agent orchestration, grounded RAG, evaluation, reliability, or AI engineering infrastructure, I would like to hear about the role.

Also open to selected agentic-systems consulting and build collaborations.