~/jaluma
// product_ai_engineer

Javier
Martínez

I build AI systems at the intersection of research and production. Hard problems with people who care about shipping things that work.

Currently leading AI at Zylon, building private AI infrastructure for enterprise customers in regulated industries.

0k+

GitHub Stars

open source

0+

Years GenAI

at Zylon

0+

Years Engineering

Software Engineering

// experience
  1. Zyloncurrent

    Product AI Engineer

    May 2024 – Present

    Lead AI product development end to end: setting technical direction, shipping features, and turning a greenfield product into a platform used by enterprise customers in regulated industries.

    PrivateGPT

    Main contributor. 57k+ ⭐ open-source backbone of the platform.

    Model serving

    Triton + vLLM on Kubernetes, fully on-premises deployments

    Sandboxed execution

    Secure runtime that lets LLMs perform real actions safely

    Agentic systems

    Tool calling, MCP, RAG. LangChain, LlamaIndex, custom framework.

  2. Freightol

    Software Engineer

    Sep. 2020 – May 2024

    • Industry 4.0 microservices on AKS with Terraform, integrating ERP and CRM systems.
    • RAG-based AI system for querying company documents through natural language.
    • AI-driven data extraction pipeline scraping and structuring external sources.
  3. Seresco

    Software Engineer Intern

    Sep. 2019 – Apr. 2020

    • AI-based automation systems for factory floor tasks using TensorFlow.
    • Microservice architectures on AKS.
// domains

Open source GenAI

Deep in the OSS AI ecosystem. Main collaborator on PrivateGPT, contributor to LlamaIndex and Ollama. Most of production AI runs on code someone shared for free.

PrivateGPTLlamaIndexOllama

Local and private inference

Deploying models on your own hardware: vLLM and Triton on Kubernetes with GPU Operator, serving LLMs, embeddings, rerankers, and OCR fully on-premises, including air-gapped environments.

vLLMTritonKubernetes

Multi-agent systems

Agents that do real work: tool calling, MCP servers, RAG over large document corpora, and sandboxed code execution. Building the infrastructure that lets LLMs act, not just respond.

LangChainMCPRAG

What I am exploring

Agent memory and long-horizon reasoning: how AI systems build, retain, and use knowledge across sessions. Evaluation frameworks, structured reasoning, and where GenAI meets classical software.

MemoryReasoningEvaluation
// open_source

LlamaIndex

run-llama/llama_index

Ollama

ollama/ollama

Contributed during early adoption while building private AI infra at Zylon.

// education
// research

PhD Research

University of Oviedo · sep. 2026

incoming

Topic: memory systems for LLM-based agents.

The question I care about: how do agents build, retain, and use knowledge across long interactions? This connects directly to production systems I work on — where agents need to plan, act, and recover across complex workflows without losing context.

MemoryLong-horizon reasoningAgent evaluationLLM agents