11/09/2026 updated
100 % available
AI Engineer | Production LLM Agents & Model Evaluation
San Cristóbal de La Laguna, Spain (Canary Islands)
Only remote
BSc in Computer EngineeringAbout me
AI engineer building production LLM agents on AWS: Bedrock AgentCore, LangGraph, MCP, Amazon Connect. Strong focus on evaluation, guardrails and reliability. English and Spanish. Part-time availability for remote projects.
Cloud EngineeringLarge Language ModelsPrompt EngineeringApi GatewayTerraform
LLM Systems & Agentic AI
Expertise in building and deploying production LLM systems including Amazon Bedrock (Claude), Bedrock AgentCore, LangGraph agents with tool calling, MCP and A2A communication, RAG pipelines (hybrid retrieval, reranking, parent-child chunking), prompt engineering, cost measurement, LLM-as-a-judge evaluation (RAGAS), Bedrock Guardrails, LoRA/QLoRA fine-tuning with Hugging Face, and vLLM serving with quantization and batching.
Cloud Infrastructure & Data Engineering
Deep hands-on experience with AWS services including Lambda, SQS, S3, Firehose, API Gateway, Amazon Connect, VPC, CloudWatch, Secrets Manager, and CDK. Proficiency in Terraform, Docker, Kubernetes, GitHub Actions, Prometheus and Grafana for production monitoring, as well as Kafka/Confluent with Avro and Schema Registry, Snowflake, PostgreSQL, Qdrant, pgvector, and Redis.
MLOps & Model Fine-Tuning
Full MLOps lifecycle management including training, evaluation, and model versioning. Deployment of services on Kubernetes with Docker, production monitoring with Prometheus and Grafana, end-to-end LLM fine-tuning workflows (LoRA/QLoRA) on proprietary datasets, and serving tuned Qwen and Mistral models in production with vLLM on Kubernetes (OVHcloud GPUs) using quantization and continuous batching.
Integration & Security
Strong knowledge of OAuth2/OIDC (Entra ID, on-behalf-of), JWT with scopes and app roles, WSO2 API Manager, OpenAPI, SigV4, TLS and firewall troubleshooting, and CloudEvents for secure API integration.
Programming Languages & Frameworks
Primary language Python, with additional proficiency in SQL, TypeScript, Go, and Java. Frameworks and tools include FastAPI, React, Node.js, PyTest, and integration tests against real services.
RAG & Retrieval Systems
Design and implementation of RAG systems with hybrid retrieval (dense + BM25 + RRF), reranking, parent-child chunking, and reproducible evaluations benchmarked with bootstrap CIs. LLM judge calibration against human labels for faithfulness assessment.
Computer Vision Pipelines
Ownership of multi-country production computer-vision pipelines processing approximately 50k installation photos per day, with prompt caching on Amazon Bedrock (Claude), achieving approximately 85% cost reduction per image and output validation catching 6-7% silent failures.
Document Processing & OCR
Design and deployment of high-resolution document-classification pipelines with locally trained models, agentic document-processing workflows with autonomous tool selection and multi-step reasoning, and OCR combined with LLM prompt engineering for structured data extraction from scanned documents.
Spec-Driven Development & ADRs
Coordination of rollouts across multiple teams using Architecture Decision Records (ADRs) and spec-driven development practices, including Scrum/Kanban methodologies and integration testing against real services.
Expertise in building and deploying production LLM systems including Amazon Bedrock (Claude), Bedrock AgentCore, LangGraph agents with tool calling, MCP and A2A communication, RAG pipelines (hybrid retrieval, reranking, parent-child chunking), prompt engineering, cost measurement, LLM-as-a-judge evaluation (RAGAS), Bedrock Guardrails, LoRA/QLoRA fine-tuning with Hugging Face, and vLLM serving with quantization and batching.
Cloud Infrastructure & Data Engineering
Deep hands-on experience with AWS services including Lambda, SQS, S3, Firehose, API Gateway, Amazon Connect, VPC, CloudWatch, Secrets Manager, and CDK. Proficiency in Terraform, Docker, Kubernetes, GitHub Actions, Prometheus and Grafana for production monitoring, as well as Kafka/Confluent with Avro and Schema Registry, Snowflake, PostgreSQL, Qdrant, pgvector, and Redis.
MLOps & Model Fine-Tuning
Full MLOps lifecycle management including training, evaluation, and model versioning. Deployment of services on Kubernetes with Docker, production monitoring with Prometheus and Grafana, end-to-end LLM fine-tuning workflows (LoRA/QLoRA) on proprietary datasets, and serving tuned Qwen and Mistral models in production with vLLM on Kubernetes (OVHcloud GPUs) using quantization and continuous batching.
Integration & Security
Strong knowledge of OAuth2/OIDC (Entra ID, on-behalf-of), JWT with scopes and app roles, WSO2 API Manager, OpenAPI, SigV4, TLS and firewall troubleshooting, and CloudEvents for secure API integration.
Programming Languages & Frameworks
Primary language Python, with additional proficiency in SQL, TypeScript, Go, and Java. Frameworks and tools include FastAPI, React, Node.js, PyTest, and integration tests against real services.
RAG & Retrieval Systems
Design and implementation of RAG systems with hybrid retrieval (dense + BM25 + RRF), reranking, parent-child chunking, and reproducible evaluations benchmarked with bootstrap CIs. LLM judge calibration against human labels for faithfulness assessment.
Computer Vision Pipelines
Ownership of multi-country production computer-vision pipelines processing approximately 50k installation photos per day, with prompt caching on Amazon Bedrock (Claude), achieving approximately 85% cost reduction per image and output validation catching 6-7% silent failures.
Document Processing & OCR
Design and deployment of high-resolution document-classification pipelines with locally trained models, agentic document-processing workflows with autonomous tool selection and multi-step reasoning, and OCR combined with LLM prompt engineering for structured data extraction from scanned documents.
Spec-Driven Development & ADRs
Coordination of rollouts across multiple teams using Architecture Decision Records (ADRs) and spec-driven development practices, including Scrum/Kanban methodologies and integration testing against real services.
Languages
EnglishFluentSpanishNative speakerTurkishBasic knowledge
Project history
Design long-horizon agentic RL environments for frontier-model training with realistic multi-step data and software tasks built on public datasets, hidden automated graders and anti-shortcut controls. Calibrate task difficulty through repeated frontier-model trials against pass-rate and step-count targets.
Own the multi-country production computer-vision pipeline (~50k installation photos/day, Claude on Amazon Bedrock), cutting cost per image by 85% with prompt caching. Redesigned it as event-driven (Lambda, SQS, S3, Firehose, Snowflake). Built LangGraph agents on Amazon Bedrock AgentCore for field operations and customer bookings. Worked on contact-center migration from Genesys to Amazon Connect integrating AI agents. Designed and shipped Kafka integration for field-technician app delivery. Defined agent exposure through corporate API Manager and coordinated rollouts across six teams.
Delivered a fully local RAG system over 50+ IRS tax-compliance documents for a US client using Ollama, Qdrant and FastAPI in Docker with zero cloud dependencies, parent-child chunking and a three-layer isolation architecture. Automated the Spanish regulatory form Modelo 790 for legal/compliance clients with n8n and LLM extraction pipelines.