09/22/2026 updated
100 % available
Principal AI Engineer | Python, LLMs, RAG & Neo4j
Phuket, Thailand
Only remote
MSc Machine Learning & Big Data (UPM); Stanford AI Professional CertificateAbout me
I build production AI and graph systems, with six years across automotive, telecom and finance. Remote from Thailand; available for onboarding, equipment collection and planned onsite visits every few months.
AI AgentsRAGLLMGCPPython (Programming Language)Machine LearningNeo4jGoogle CloudFlaskDeep LearningGenerative AIFastAPIRecurrent Neural Networks
I’m a Principal AI Engineer and co-founder and sole engineer of PAI, with six years of software engineering experience across automotive, telecom and finance. I build production AI and data systems, taking responsibility from architecture and implementation through deployment and testing.
LLM applications and AI agents
Built PAI, a multichannel AI assistant spanning web, WhatsApp and Telegram, with more than 85 tools for documents, finances, calendars and collaboration. My work covers tool-calling agents, RAG, hybrid retrieval using pgvector and BM25, embeddings, multimodal OCR, evaluation and inference-cost controls. I implement bounded agent execution, confirmation controls for destructive actions, and ledger-backed undo for supported creation actions.
Graph machine learning and anomaly detection
Build graph ML systems for German automotive engineering teams, including anomaly detection and matching signals across ECU software versions. Experience includes PyTorch, PyTorch Geometric, GATv2, GraphSAGE and representation learning on graphs with over one million edges. I evaluate models using synthetic mutations and held-out ECU families, including AUROC and AUPRC analysis.
Neo4j and graph-performance engineering
Extensive experience with Neo4j, Cypher, reachability, transitive closure and graph comparison. I use techniques including CSR adjacency and bitset operations to improve performance. At Concentrio, I reduced graph-reachability workloads from more than 12 hours to approximately 20–27 seconds.
Backend systems, data and deployment
Develop services with Python, FastAPI, Flask and Java. My work includes asynchronous processing with RabbitMQ, PostgreSQL with Row-Level Security, MongoDB, and integrations between heterogeneous data sources. I deploy systems using Google Cloud Run, Cloud SQL, Cloud Storage, Firebase and Docker, with React interfaces and automated testing through pytest and Playwright.
Additional experience
Developed reinforcement-learning and genetic-algorithm strategies for financial applications, alongside sentiment-analysis pipelines. I also taught a 32-hour deep-learning module covering neural-network fundamentals, training dynamics and practical implementation.
Experience
Professional experience5 - 10 years
Freelancer since2026
Languages
EnglishFluentSpanishNative speaker
Project history
Built a production graph-ML platform for automotive ECU analysis, supporting anomaly detection and comparison of signals across software versions on graphs with over one million edges. Implemented GNN models using PyTorch Geometric, GATv2 and GraphSAGE, with evaluation through synthetic mutations across eight held-out ECU families. Delivered the platform through FastAPI and interactive analysis reports, reducing analysis times from more than 12 hours to 10–60 seconds.
Co-founded PAI and built its software platform end to end: a shared AI assistant across web, WhatsApp and Telegram, with over 85 tools for documents, finances, calendars and collaboration. Deployed and in daily use ahead of public launch.
Implemented hybrid retrieval with pgvector and BM25, bounded agent execution, confirmation controls for destructive actions, and ledger-backed undo for supported creation actions. Built the backend, React interfaces, integrations and cloud deployment, including PostgreSQL Row-Level Security, usage-based spending controls, OCR and approximately 1,200 automated tests.
Taught the 32-hour Deep Learning module of KeepCoding's Big Data, AI & Machine Learning bootcamp (17th edition), covering neural-network fundamentals, training dynamics, and modern architectures for cohorts of career-transitioning professionals.