Guilherme Inácio, AI Engineer

AI Engineer building agent infrastructure, MCP servers and LLM applications.

Guilherme Inácio is a Computer Engineer working as AI Engineer at Abstrato, based in Florianópolis, Santa Catarina, Brazil (America/Sao_Paulo). Last updated 2026-08-21.

I am Guilherme Inácio, a Computer Engineer (UFSC) working as an AI Engineer at Abstrato, where I build agentic architectures and AI infrastructure for enterprise clients. My work centres on AI agents, the Model Context Protocol (MCP), retrieval systems and the data platforms that support them. Before Abstrato I spent two years at the fintech Franq as an Applied AI Engineer and Data Analyst, shipping document-analysis agents, LLM observability and internal AI tooling. Earlier I worked in business intelligence at Meton Soluções and researched EEG signal classification for epilepsy surgery at the Czech Technical University in Prague.

This site is agent-native: an AI agent can read this CV and book a meeting through the Model Context Protocol endpoint at https://guinacio.cv/mcp, the REST API under /api/agent, or the in-page WebMCP tools. No scraping required.

Work experience

Roles in reverse-chronological order, with the responsibilities and delivered work of each.

AI Engineer, Abstrato

July 2026 - Present (current role)

  • AI Agent Development (LangChain, LangGraph, Pydantic AI, Claude Agent SDK)
  • Agent Infrastructure & Deployment (FastMCP, FastAPI, Docker)

Applied AI Engineer, Franq

June 2025 - July 2026

  • AI-Powered Data Apps Development (Streamlit, Docker)
  • AI Agent Systems & Infrastructure (LangGraph orchestration, Claude Agent SDK, API deployment, LangServe, VertexAI, MCP, Docker)
  • Data Analysis (Python / BigQuery / DBT / GCP)

Automated Validation & Document Analysis System

  • Built automated data extraction and validation for financial documents, removing most of the manual verification workload
  • Built automated model evaluation pipeline comparing document analysis outputs against gold-standard datasets with drift detection using domain-specific metrics (approval/recusal rates, tag-level accuracy)
  • Implemented LLM observability with LangFuse for robust monitoring of AI agent performance in production

Structured Extraction of Financial Data Documents

  • Agent system that transforms unstructured financial documents into structured, queryable data
  • Extracts and normalizes financial information for downstream analytics and modeling
  • Opened up structured search over large volumes of documents that were previously manual-only

AI-Enabled Internal Productivity Apps

  • Built AI-powered internal tools for teams across the company
  • Introduced Streamlit for internal applications and drove its adoption company-wide

Data Analyst, Franq

October 2024 - June 2025

  • Data Analysis (Python / BigQuery / DBT / GCP)
  • Development of Data Apps (Streamlit)
  • AI/LLM Development (VertexAI, LangChain, LangGraph, MCP)

Developed comprehensive Incentive Campaign Platform

  • Centralized performance monitoring for CS, Sales, and other teams
  • Integrated Google Cloud services (BigQuery analysis, Vertex AI rule querying)
  • Built administrative tools and configuration management
  • Automated complex campaign closing calculations and reporting
  • Improved efficiency and data visibility across the company

Business Intelligence Analyst, Meton Soluções

March 2022 - October 2024

  • Dashboard creation (Streamlit / PowerBI / BoldBI)
  • ETL routines (Pentaho Spoon)
  • Database maintenance / optimization (PostgreSQL)
  • Data Analysis (Python)

Dashboard system for commission calculations across 50 dealership groups

  • Implemented flexible commission structures based on dealership, item, and group specifications
  • Automated calculations for sales representatives and managers
  • Integrated complex commission rules with sales data for accurate payment processing

Database optimization and maintenance

  • Achieved 300%+ query speed improvement through index optimization
  • Made a data-intensive client accessible for the first time
  • Enhanced efficiency through normalization and redundancy removal

Data Analysis Intern, Czech Technical University in Prague (CTU)

January 2019 - March 2019

  • Statistical analysis of EEG data for research on improving epilepsy surgical outcomes.
  • Continued the work remotely from Brazil after the internship, presenting the findings as the final undergraduate project.
  • Applied machine learning to EEG classification as part of the ongoing research.

Skills

Self-assessed proficiency from 0 to 100, alongside years of professional experience and the number of projects each technology has been used on.

Programming Languages

  • Python: proficiency95 of 100, 6 years of professional experience, used on 20 projects
  • SQL: proficiency90 of 100, 5 years of professional experience, used on 18 projects
  • C: proficiency60 of 100, 1 year of professional experience, used on 2 projects
  • Java: proficiency65 of 100, 1 year of professional experience, used on 1 projects
  • Lisp: proficiency60 of 100, 1 year of professional experience, used on 2 projects

AI/ML

  • LangChain: proficiency93 of 100, 2 years of professional experience, used on 10 projects
  • LangGraph: proficiency88 of 100, 1 year of professional experience, used on 7 projects
  • PyTorch: proficiency50 of 100, 1 year of professional experience, used on 1 projects
  • Scikit-learn: proficiency60 of 100, 1 year of professional experience, used on 1 projects
  • Statsmodels: proficiency75 of 100, 3 years of professional experience, used on 8 projects
  • MCP: proficiency90 of 100, 1 year of professional experience, used on 10+ projects
  • FastMCP: proficiency88 of 100, 1 year of professional experience, used on 5 projects
  • Pydantic AI: proficiency70 of 100, 1 year of professional experience, used on 1 projects
  • Claude Agent SDK: proficiency90 of 100, 1 year of professional experience, used on 3 projects

Data Science

  • NumPy: proficiency90 of 100, 5 years of professional experience, used on 22 projects
  • Pandas: proficiency92 of 100, 5 years of professional experience, used on 25 projects
  • Matplotlib: proficiency85 of 100, 4 years of professional experience, used on 20 projects
  • Seaborn: proficiency80 of 100, 3 years of professional experience, used on 15 projects
  • Plotly: proficiency85 of 100, 2 years of professional experience, used on 10 projects

Tools & Platforms

  • Streamlit: proficiency90 of 100, 3 years of professional experience, used on 15 projects
  • PostgreSQL: proficiency85 of 100, 4 years of professional experience, used on 12 projects
  • BigQuery: proficiency80 of 100, 2 years of professional experience, used on 8 projects
  • Docker: proficiency75 of 100, 2 years of professional experience, used on 6 projects
  • Git: proficiency90 of 100, 6 years of professional experience, used on 25 projects
  • FastAPI: proficiency85 of 100, 1 year of professional experience, used on 4 projects
  • pgvector: proficiency80 of 100, 1 year of professional experience, used on 2 projects

Featured projects

Open-source and private work, mostly around the Model Context Protocol, AI agents and data applications. Source code is on GitHub unless a project is marked private.

LangChain MCP Client

Streamlit playground for connecting any model to any MCP server. Supports OpenAI, Anthropic and Google, with real-time chat, server management and tool interactions.

Category: AI/ML. Primary language: Python. Stack: Python, Streamlit, LangChain, MCP, OpenAI, Anthropic.

Streamlit Portal Server

Application portal with authentication, user management, app discovery and admin controls for running multiple Streamlit apps in production.

Category: Web Development. Primary language: Python. Stack: Python, Streamlit, Authentication, Docker, Admin Panel.

MCP Google Calendar

MCP server that gives Claude and other AI assistants full access to Google Calendar: event management, scheduling and availability checking.

Category: AI/ML. Primary language: Python. Stack: Python, MCP, Google API, Calendar Integration, AI Tools.

MCP Google Workspace

Production-ready Google Workspace MCP server: Gmail, Calendar, Drive, Sheets, Docs, Tasks, People, Forms, and Slides in one composed FastMCP package with resources, prompts and sampling.

Category: AI/ML. Primary language: Python. Stack: Python, MCP, FastMCP, Google API, Gmail, Google Drive, AI Tools.

Better Memory MCP

MCP memory server with graph operations: observation-level search, relation filtering, neighbor traversal, subgraph extraction and path finding. Fork of @modelcontextprotocol/memory with a wider query surface.

Category: AI/ML. Primary language: TypeScript. Stack: Python, MCP, Graph, Search, Filtering, Neighbor Traversal, Subgraph Extraction, Path Finding.

Claude Image Gen

Image generation with Google's Nano Banana, wired into Claude Code. Runs as a CLI script or an MCP server, with configurable aspect ratios and model selection for the quality/speed trade-off.

Category: AI/ML. Primary language: TypeScript. Stack: Node.js, TypeScript, Nano Banana, MCP, Claude Skills, Claude Code.

SRAG Analytics

Healthcare analytics over SRAG (Severe Acute Respiratory Syndrome) data from DATASUS. LangGraph agents orchestrate SQL queries, news retrieval, RAG over the data dictionary and report generation, with safety guardrails, audit trails and Portuguese language support.

Category: Data Analysis, AI/ML. Primary language: Python. Stack: Python, LangGraph, OpenAI, FastAPI, Streamlit, PostgreSQL, Docker, Tavily API, pgvector.

Sentinel Assistant

Autonomous AI assistant for Windows and macOS. A persistent background agent monitors email and calendar, runs hourly heartbeats and browses the web with human-in-the-loop approval, surfaced through a native pywebview chat GUI with inbox, reminders, TODOs and session history.

Category: AI/ML. Primary language: Python. Stack: Python, Claude Agent SDK, MCP, pywebview, RAG/KG, SQLite.

MCP Microsoft

Microsoft 365 MCP server reaching six services (Mail, Calendar, OneDrive, SharePoint, Contacts and Teams) with read, write and action support across all of them. Handles personal and enterprise accounts with multi-account profiles, and ships as an MCPB bundle for Claude Desktop.

Category: AI/ML. Primary language: Python. Stack: Python, MCP, FastMCP, Microsoft Graph API, MSAL, MCPB, AI Tools.

Other projects

Infinite AI Backroom

Multi-persona AI conversation platform on Ollama. Define personas with distinct roles, @mention them in a shared thread, and review the conversation through built-in log analysis.

Category: AI/ML. Primary language: Python. Stack: Python, Streamlit, Ollama, LLM, NLP, AI Agents.

Claude Streamlit Agents

Specialized AI agents for Streamlit apps: UI/UX design, performance optimization, security auditing and other domains, each with its own expertise.

Category: AI/ML. Primary language: Python. Stack: Python, Streamlit, Claude AI, MCP, Agent Architecture, AI Tools.

Agent Forge

GUI workbench on top of Claude Code and the Agent SDK for creating AI workflows and agents on demand.

Category: AI/ML. Primary language: TypeScript. Stack: TypeScript, Next.js, Claude Agent SDK, Claude Code.

  • Private repository

Database Chat Agent

Chat interface for querying databases in natural language, with text and voice input, automatic SQL generation and data visualization.

Category: AI/ML. Primary language: Python. Stack: Python, Streamlit, LangGraph, Google Gemini, SQLite, matplotlib.

  • Private repository

Kill Bill Landing

Landing page built around the Kill Bill theme.

Category: Web Development. Primary language: TypeScript. Stack: TypeScript, React, Vite.

Suburban Schemers

Landing page for the movie Fun with Dick and Jane.

Category: Web Development. Primary language: TypeScript. Stack: TypeScript, React, Vite.

Guardians Music Player

3D music player inspired by Guardians of the Galaxy.

Category: Web Development. Primary language: TypeScript. Stack: TypeScript, React, Vite, Three.js.

Education

  • Bachelor's in Computer Engineering, Federal University of Santa Catarina (UFSC), 2014 - 2019 (Recognized in Germany by ZAB - Zentralstelle für ausländisches Bildungswesen)
  • Exchange Program - Informatics, Technische Universität München (TUM), 2015 - 2016 (2 semesters)
  • Data Analyst with Python, DataCamp (66h). Certificate

Contact

Email is the fastest way to get a reply. To meet, pick a time at /book, or let an AI agent do it through the Model Context Protocol endpoint. Either way the calendar is checked live, and every booking is confirmed by the attendee over email before anything reaches the calendar.