I build software that connects AI models, tools, APIs, databases and real workflows.
> whoami
Luis Carlos Fuentes De Avila
> profile
Systems Engineer focused on Applied AI, Agentic Systems, MCP tooling,
backend platforms, data automation and open-source collaboration.
> current_signal
LLM applications · RAG systems · AI agents · backend APIs · real dataI am a Systems Engineer focused on building practical software with Applied AI.
My current work is centered on:
- building LLM-powered applications;
- designing RAG systems that use real documents and real data;
- creating AI agents that use tools safely;
- working with MCP tooling and agent workflows;
- building backend systems with APIs, databases, tests and documentation;
- turning AI ideas into usable software products.
I also have a Machine Learning and Deep Learning foundation through Andrew Ng / DeepLearning.AI, including model evaluation, neural networks, regularization, hyperparameter tuning and optimization.
|
LLM apps, RAG systems, AI agents and automation workflows using real data. |
Backend, APIs, databases, tests, Docker, TypeScript, Python and maintainable systems. |
MCP tooling, tool-using agents, guardrails, workflow automation and AI safety patterns. |
|
Open-source work around AI governance, MCP tool trust, LlamaIndex guards and deterministic policy enforcement for AI agents. Why it matters: shows agent safety, MCP thinking, guardrails, policy design and real open-source contribution. 
|
Contributions to computer-vision / ML tooling: ONNX runtime paths, TPS pipeline work, MediaPipe landmark wrappers and regression tests. Why it matters: shows Python, ML pipelines, testing discipline and ability to work inside complex codebases. 
|
|
Async DNS benchmark API with safe sync/async behavior, fallback consistency and tests for a Python networking tool. Why it matters: shows Python library work, async programming, API stability and production-minded behavior. 
|
Small but clean contribution to a real Python SDK: return type hint improvement for better type checking and IDE support. Why it matters: shows comfort with SDK contribution flow, scoped PRs, code quality and open-source standards. 
|
|
End-to-end data engineering project: ingestion, DuckDB warehouse, dbt models, data tests, reconciliation and API serving. Why it matters: shows data engineering, SQL modeling, quality gates and reproducible analytics pipelines. 
|
Private product-style platform with API, MCP server, shared contracts and conversational architecture. Why it matters: even if private, it represents backend architecture, MCP direction, product thinking and applied AI systems design. 
|
I use open source to practice real engineering: reading existing code, understanding maintainers' expectations, writing scoped PRs, adding tests and documenting behavior.
-
AI governance and agent safety
- MCP tool trust policy.
- LlamaIndex tool guard integration.
- Safety policies for AI agent execution.
-
Python libraries and tooling
- Async APIs.
- Type hints.
- CLI behavior.
- Runtime stability.
- Tests and backward compatibility.
-
ML / computer vision infrastructure
- TPS / RBF pipeline work.
- ONNX runtime integration.
- MediaPipe landmark wrappers.
- Regression and parity tests.
2026 focus:
[01] Build production-style LLM applications
[02] Build RAG systems with evaluation and citations
[03] Build MCP tools and agent workflows
[04] Strengthen backend architecture with TypeScript and Python
[05] Keep contributing to open source with scoped, tested PRsReadable code > clever code
Small PRs > giant rewrites
Real tests > optimistic assumptions
Clear contracts > hidden behavior
Useful AI > flashy demosBuilding software where AI is not decoration — it is part of the system.