AI engineer · Agent systems
I’m Jancarlos Sosa, an AI engineer focused on agentic workflows, applied intelligence, and the product engineering required to make them dependable.
Reliable agents · human control · useful automation
Operating principle / 02
My work sits between model capability and operational reality: permissions, APIs, context, failure modes, and the moment a person should remain in control.
The goal is not autonomous software for its own sake. It is a system that removes repetition, exposes its reasoning, and earns the right to act.
Selected systems / 03
AI-native personal finance
A production finance platform with a guarded MCP interface for natural-language access to real financial data.
Three independently deployable services, one permission-aware system.
Voice, retrieval, and documents
An AI invoicing workflow that turns natural speech and historical documents into structured, editable invoices.
Speech-to-text, retrieval, extraction, and PDF generation in one workflow.
Multi-account trading operations
A focused trading utility that routes one decision across multiple authenticated brokerage accounts.
Reduced a repetitive multi-account workflow from roughly 50 minutes to seconds.
Field record / 04
Designing agentic systems for enterprise workflows where perception, automation, and human review must work together reliably.
MCP, computer use, tool orchestration, human-in-the-loop controls
RAG, multimodal models, OCR, speech, evaluation-minded workflows
Python, Java, TypeScript, FastAPI, Spring Boot, React, Next.js
AWS, PostgreSQL, Supabase, Docker, APIs, authentication
Working notes / 05
The notebook is open. The first field note is being prepared.
New writing will appear here.Open channel / 06
I’m interested in applied AI roles, thoughtful collaborations, and systems where reliability matters as much as novelty.