LLM & Applied AI
Building model-backed features as software systems rather than isolated prompts.
LLM workflows Structured outputs Tool calling Prompt design Model evaluation Model routing
Professional Profile
Marcos García Estévez
AI / GenAI Engineer
AI / GenAI Engineer|
Positioning
About Me
I build applied AI systems around large language models, retrieval and agent workflows, with an emphasis on turning model capabilities into reliable software: grounded assistants, data integrations, evaluation loops, safety controls and operational tooling.
I currently work as a Full Stack LLM Development Analyst at Accenture Spain. Alongside enterprise work, I build and operate projects spanning production RAG, automated product pipelines and policy-aware agent tooling. My strongest area is the engineering layer around AI: APIs, automation, evaluation, observability, testing and controlled releases.
Current role
Full Stack LLM Development Analyst
Focus
LLM systems, RAG & AI agents
Location
Málaga, Spain
Languages
Spanish · English (B1 Cambridge)
Core capabilities
Engineering Focus
Retrieval & RAG
Grounding answers on controlled knowledge and measuring retrieval quality.
Hybrid retrieval Embeddings Reranking Chunking Semantic search RAG evaluation
Agents & Safety
Agent orchestration with explicit permissions, evidence and fail-safe behaviour.
LangGraph MCP Agent workflows Policy controls Human-in-the-loop Safety & abuse controls
Engineering & Operations
Backend, persistence and operational practices that keep AI features maintainable.
Python APIs Cloud services CI/CD Observability Testing Docker & Linux
Technologies used across professional and public work
Python LangChain LangGraph LlamaIndex FastAPI Flask FastMCP / MCP OpenAI API Gemini Transformers BERT MLflow scikit-learn Pandas AWS Lambda Docker Linux DuckDB SQLite SQL TypeScript GitHub Actions
Professional track record
Professional Experience
Full Stack LLM Development Analyst
Accenture España · Full-time
June 2025 - Present · Málaga, Andalucía, Spain · Hybrid
- Design and develop AI/LLM workflows for business processes, with attention to traceability, maintainability and operating cost.
- Integrate and expose internal data through APIs and cloud services, including data-driven recommendation and decision-support flows.
- Build reusable components for automation, validation and operation of AI solutions instead of one-off model integrations.
- Work across the boundary between backend, data and model orchestration using Python, AWS Lambda, LangChain, LangGraph and LLM APIs.
Python LangChain LangGraph AWS Lambda LLM APIs APIs Data Workflows
Progressed from the LLM internship into the full-time analyst role
Full Stack LLM Intern
Accenture España · Internship
March 2025 - May 2025 · 3 months · Málaga, Andalucía, Spain · On-site
- Adjusted and evaluated BERT-based models for multilingual text classification and NLP tasks.
- Built automated model-comparison workflows and used MLflow to track experiments, metrics and results.
- Contributed to machine-learning workflows for detecting problematic content and evaluating model behaviour.
Python BERT Transformers NLP MLflow Model Evaluation OpenAI API
LLM Quality Specialist - Spanish
Outlier · Freelance
May 2023 - July 2024 · 1 year 3 months · Remote
- Systematically evaluated Spanish LLM responses for factual accuracy, relevance, instruction following, reasoning, inconsistencies and hallucinations.
- Reviewed training and evaluation outputs and produced structured feedback to improve model quality and end-user usefulness.
- Worked with detailed evaluation guidelines while maintaining linguistic consistency and quality across repeated comparisons.
LLM Evaluation AI Quality NLP Spanish
WordPress Developer & Cloud Administrator
GrupoOro · Internship
March 2024 - June 2024 · 4 months · Málaga, Spain · Remote
- Administered Linux VPS environments and Plesk, applying updates, security measures and log-based troubleshooting.
- Configured Cloudflare services including DNS, CDN, WAF and Access, and developed and optimized WordPress sites.
Linux Cloudflare Plesk WordPress DNS WAF
Selected engineering work
Production & Technical Projects
Three projects that best represent my current direction: production conversational AI, retrieval/context infrastructure and safer agent execution.
Production-oriented AI product · Ongoing
Awaita (Awi)
Private AI guide for Menorca. My contribution is focused on the conversational product layer: grounded answers, retrieval quality, evaluation, safety, observability and controlled releases.
- Develop and tune the retrieval workflow around curated sources, including comparative evaluation and reranking decisions.
- Connect feedback, analytics, latency, cost and quality signals to improve real conversations with evidence.
- Maintain safety and abuse controls plus staged CI/CD checks for reliable releases to staging and production.
Conversational AI RAG Retrieval Evaluation Reranking AI Safety Observability CI/CD
Live automated product · Production
SoloGangas
Production deal-discovery platform with a public website and a 2.3K+ Telegram audience, operated end to end as an automated, data-backed product across both surfaces.
- End-to-end automation handles content processing, publication, updates and lifecycle management.
- Production engineering built on durable persistence, separated runtime services, monitoring, backups and controlled recovery paths.
- Staged CI/CD validates changes before deployment, with health checks and rollback.
Python PostgreSQL Flask AI-assisted Automation Docker CI/CD Observability
Agent safety tooling · August 2026 · Ongoing
OpenCode Permission Reviewer
Unofficial OpenCode community plugin that reviews permission prompts with a tool-free model session and policy-aware evidence, while keeping hard safety invariants in deterministic code.
- Allows once, denies with feedback or escalates to a human; critical risk and uncertain failures cannot silently auto-approve.
- Builds bounded, redacted evidence from user intent, transcript context and optional read-only Git/SSH/script enrichment.
- Adds audit records, human-override race handling, a TUI status layer and CI covering formatting, lint, types, tests and build.
TypeScript Bun OpenCode Agent Safety Policy Engine Human-in-the-loop GitHub Actions
Background
Education & Training
Advanced Vocational Specialization in Artificial Intelligence & Big Data
CPIFP Alan Turing · Dual program with Accenture
September 2024 - June 2025
Spanish Máster FP / Curso de Especialización: a 600-hour dual vocational program focused on AI, machine learning, Python, Big Data, NLP and cloud. Final grade: 10/10. Capstone: MIDAS, a modular multi-agent platform for automating data-science and machine-learning workflows.
Linux & Cloud Bootcamp
EducacionIT
June 2024 - August 2025
Program covering Linux administration, networking, Bash, databases and SQL, cybersecurity fundamentals, hosting, cloud computing and containers.
Higher Vocational Training in Web Application Development (DAW)
MEDAC
September 2022 - June 2024
Practice-oriented software development program covering web applications, databases, backend/frontend development, deployment and software engineering fundamentals.
Software Engineering studies · 3 academic years completed
University of Málaga
September 2019 - June 2022
Completed three academic years of the Software Engineering degree before moving to a more applied vocational path. Degree not completed.
Professional contact
Open to the right AI engineering opportunity
For professional opportunities or technical collaboration, LinkedIn is the best way to reach me. For a deeper look at implementation details, Warcos Labs and GitHub contain my public engineering work.