AI Engineering competencies.
Eleven competencies, each with what I applied, in which project and in what status. The decisions and their trade-offs, without inflating what is still in construction.
AI agents and tool calling
AgendaGo In production · Nux (Bonuxo) In constructionModels that choose which tool to call, and with which parameters, to operate a real system.
Multi-agent orchestration with LangGraph
AI agents platform for companies In constructionAgent graphs with a router, SQL and RAG subgraphs, bounded retries and memory with a checkpointer.
LangChain
AI agents platform for companies In constructionInterchangeable models, tools, loaders, splitters and PGVector integration.
RAG and semantic search
AI agents platform for companies In constructionEmbeddings, chunking and pgvector to answer with a document and page citation.
Safe text-to-SQL
AI agents platform for companies In constructionNatural-language queries over databases: validated SELECT only, a view allowlist, a read-only user and per-tenant RLS.
LLM evaluation
AgendaGo In productionOwn datasets (100 phrases; 52 scenarios and 15 WhatsApp conversations) and per-call cost and latency telemetry.
AI guardrails and safety
AgendaGo In production · AI agents platform for companies In constructionInput and output filters, prompt injection defense and multi-tenant isolation.
Observability
AgendaGo In production · AI agents platform for companies In constructionTraces with Langfuse and per-call usage logging.
Integrating LLMs with business systems
AgendaGo In production · Bonuxo In productionConnecting models to the WhatsApp Cloud API, MercadoPago and POS APIs.
Vision with multimodal models
AgendaGo In productionReading supplier invoices and delivery notes with multimodal models.
MCP (Model Context Protocol)
Bonuxo In productionOwn MCP server.