AI engineering lead · Madrid · remote EU
Retrieval, agent runtimes, guardrails, evals, traces, and the screens operators need when an answer is wrong, incomplete, or expensive. Professional software work since 2006. Machine learning since 2015.
01 / background
Professional software work since 2006, across salaried company roles and selected freelance/client work. Machine learning since 2015. Before the current LLM wave, I shipped systems for banks and retailers where defects hit money movement, customer data, compliance review, and release windows.
2006
Professional software work across product, frontend, backend, and platform systems.
2015
Retrieval, agent runtimes, guardrails, evals, traces, and operator screens.
At Atlax360 I lead AI engineering across BIFROST, ORVIAN, and Polaris: document intelligence, an AI workflow runtime, and an internal assistant. They run inside financial operations, so retrieval quality, permissions, cost, latency, and review paths matter more than a polished chat response.
I use coding agents and language models heavily in my own work. The useful part is not the fluent output. It is the surrounding system: context, tool limits, tests, traces, approvals, and clear stopping points when evidence is weak.
BBVA, Santander/PagoFX, Bankinter, Decathlon, El Corte Inglés, and current Atlax360 work.
Led architecture and delivery across frontend, backend, platform, and AI systems.
Open-source repos for agent policy, memory, tracing, logging, and local systems tooling.
02 / current work
The implementations are private. The shape is public enough to explain the work: document ingestion and retrieval, a multi-tenant AI workflow runtime, and an internal assistant with guardrails and citations.
document intelligence
Document intelligence for the documents finance actually runs on: ingestion quality gates, semantic and visual chunking, pgvector/HNSW search, caching, source-quality scoring, analytics, and explicit no-answer paths when evidence is weak.
workflow runtime
Multi-tenant AI workflow runtime: context assembly, durable memory, deterministic and cached execution tiers, queue processing, idempotency, run events, and human-review metadata when automation should stop.
internal assistant
Internal assistant that combines BIFROST retrieval with guardrails, citations, streaming UX, suggestion revalidation, and operator analytics. One surface for support, sales, and product teams.
03 / open source
Public work I can point to directly: agent policy, traces, governed memory, structured observability, local systems tooling, and developer environments.
Deterministic policy engine for AI coding agents: maps tool calls to capabilities, evaluates YAML rules, and signs allow, deny, and ask decisions in an audit log.
Self-inspecting, auditable memory for AI agents: surfaces evidence, provenance, and health behind each recall, with an optional Ed25519-signed tamper-evident ledger. Local-first, Rust.
Local-first trace recorder for AI agent runs: append-only run evidence for what the agent called, what it was allowed, and what failed, with hook ingestion for Codex/OMX harnesses.
Runtime-agnostic structured logging with automatic PII sanitization (GDPR/HIPAA/PCI-DSS) and native W3C tracing. Zero dependencies; runs on Node, Bun, Deno, Edge, and the browser.
Research harness exploring whether a coding-agent harness can improve itself through typed, layered modifications validated against operator-defined evals within strict budgets.
Zero-dependency Apple Silicon hardware monitor and thermal watchdog in Rust: CPU/GPU/ANE power, per-core load, temperatures, fans, and battery without sudo.
Open terminal environment for Ghostty on Linux: themes, presets, fonts, tmux integration, and a config switcher for a Warp-like workflow.
04 / what i work on
Most of the value sits between a source document and the person using the result. That path includes indexing, permissions, runtime state, evals, guardrails, latency, and the UI for failure handling.
Chunking, metadata, permissions, source quality, citations, and caching, with explicit refusal when retrieved sources are weak, contradictory, missing, or out of scope.
Tool boundaries, stopping conditions, traces, handoffs, evaluators, queues, and cost control, defined before an agent reaches production.
Eval sets, abstain logic, and regression traces. The discipline of checking whether a change helped before treating it as an improvement.
Permissions, audit trails, PII handling, and human-in-the-loop review for AI running inside regulated finance, where a fluent wrong answer is still a defect.
Repository context, executable plans, browser checks, CI gates, and review logs around coding agents, so their output stays verifiable inside a real codebase.
Next.js, TypeScript, streaming UX, and operator screens for model failures, showing retrieved evidence, failure state, audit trail, escalation path, and the next valid action.
05 / how i work
I work async and stay accountable for retrieval quality, runtime behavior, latency, cost, permissions, and operator workflows. Madrid-based, remote-first across the EU. Open to Staff, Principal, Architect, and Forward Deployed roles, and to contract or fractional engagements where retrieval, agents, evals, and guardrails are part of the shipped product.
06 / questions
Quick facts on stack, roles, and fit. The CV has the full timeline.
Download CV07
I work with teams shipping LLM features that need owned retrieval quality, cost ceilings, latency targets, permission checks, and failure review. Available full-time, contract, or fractional. Madrid-based, remote across the EU.