Skip to content

AI and autonomous agents

AI that does the work, not AI that demos well.

I build generative AI into processes that already exist and already matter: retrieval over your own documents, agents that can act, and the plumbing that makes both safe to run in front of a regulator.

TypeScript
Python
Java
React
Next.js
Node.js
AWS
Docker
Kubernetes
MongoDB
PostgreSQL
TensorFlow
PyTorch
Jest
Cypress
Jira
Git
GraphQL
REST
TypeScript
Python
Java
React
Next.js
Node.js
AWS
Docker
Kubernetes
MongoDB
PostgreSQL
TensorFlow
PyTorch
Jest
Cypress
Jira
Git
GraphQL
REST

RAG over your own documents

Semantic retrieval with pgvector and Elasticsearch, built so answers cite their source and a wrong answer can be traced back to the document that caused it.

Autonomous agents and MCP

Model Context Protocol servers that expose your systems as tools an agent can actually operate, with the permission boundaries drawn before the agent gets the keys.

LLM orchestration

Multi-step chains with the failure modes handled: retries, fallbacks, cost ceilings, and evaluation sets so a prompt change is measured rather than guessed at.

Integration into the real process

The model is the easy part. The work is the API contract, the queue, the audit trail and the handoff back to a person. That is what I ship.

Where I have shipped this

  • Autonomous agent wallets

    Bitcoin SV Association

    An MCP server that lets AI agents hold their own wallets and operate autonomously on-chain, plus open source libraries that automate full blockchain operations for other teams.

    API-first across multiple MCP servers · Every product controllable by an agent

  • Public sector AI

    DGT, Metro de Madrid and Madrid City Council, at Accenture

    Three years leading AI applied to data analysis for major public sector clients: predictive models, generative AI embedded in client processes, and RAG systems with semantic retrieval.

    Optimisation models for Metro de Madrid night routes · AI for automated recruitment screening

  • QRma

    Open

    Own product, in production

    Permanent QR codes for restaurants. The code engraved on the physical piece never changes; the destination behind it is edited in two taps.

    50-operation REST contract, 15 data models · MCP server exposing 14 tools

  • BrokerAI

    Proof of concept for a private banking institution

    An AI assistant that analyses and manages long-term investment fund portfolios: quantitative fund analysis, risk and return assessment with predictive models, and selection and rebalancing algorithms.

    Applied across roughly ten real portfolios · Still advising the institution occasionally

Tell me what is slow, manual, or risky.

I read every message myself and reply within one working day.

Book a call