c
Résumé
Caio Goia

Caio Goia

BI Analyst · Data modeling · Power BI

GitHub LinkedIn Email
Base
Curitiba, BR
Local time
--:-- BRT
Languages
PT · EN

BI analyst at INTECSO, working end to end on agribusiness data — from ERP and API extraction to dimensional models and Power BI. Before that, Tableau and Oracle SQL at Coamo, Paraná's largest cooperative.

What I care about most is that the number on the dashboard is right, and that nobody has to rebuild it by hand next month. That usually means replacing a manual export with a direct integration, or finding out why a refresh takes fifteen minutes.

20+Power BI dashboards in production
11industrial plants served with Tableau
−46%semantic model size, same metrics
15m → 10srefresh time after a Power Query fix

Selected work

2025 — 2026
yield error, pp — trend vs model −40% severe moderate normal good very good trendmodel

Safra Risk Radar

live

Crop-failure detector for Brazilian soybean and second-crop corn. Public data from CONAB, IBGE and NASA POWER, transformed in dbt (85 models, 78 tests) on DuckDB and BigQuery, rebuilt weekly by CI and served on a live dashboard.

Result40% less error in severe shortfall seasons than the trend baseline — and the write-up says plainly where the model doesn't help.

Python · dbt · BigQuery · DuckDB · Streamlit · GitHub Actions

Live dashboard ↗ Code ↗
Delivery Center Power BI sales overview page

Delivery Center

Power BI

Operations dashboard for a delivery platform: star schema over 369,000 orders, 46 DAX measures and a full visual redesign prototyped in Figma.

ResultA negative logistics margin on 107 of 120 days (−R$ 356k), and one hub taking 39% of total delivery time on just 2.8% of the volume.

Power BI · DAX · Power Query · Star schema · Figma

Case study ↗ Figma prototype ↗
out-of-sample AUC by target, 5-min chance 0.50 0.52 0.60 0.77 0.65 0.65 direction magnitude volatility abs. move regime

Predictability ≠ Profitability

research

A research framework that tries hard to falsify "machine learning can predict Bitcoin": 250k one-minute candles, purged cross-validation, explicit trading costs and the Deflated Sharpe Ratio. Reproducible, with CI.

ResultDirection is a coin flip (AUC ≈ 0.50). Volatility looks predictable (AUC ≈ 0.79) — until a one-line persistence baseline beats the XGBoost.

Python · XGBoost · SQLite · Statistics · CI

Code ↗
house prices vs incomes — illustrative 2015 = 100 priceincomethe exceptions

Priced Out

in progress

Housing affordability across ~50 countries, 1995–2026: a Power BI report on how house prices came apart from incomes — and where they didn't. Pulls straight from the OECD SDMX API through Power Query, so it refreshes against the live release.

StatusReport in design. The source is validated and the method note comes first: these are indices, so the report talks about change, never absolute cost.

Power BI · Power Query · DAX · OECD API

Code ↗
Also on GitHub See all →

Path

2022 — now
Experience

INTECSO

Curitiba, BR · on-site

Agribusiness R&D and technical analysis

  1. Mid-level BI Analyst

    May 2026 — present

    • Build and maintain 20+ corporate Power BI dashboards for finance, sales, logistics and laboratory teams, modeled over the TOTVS Protheus ERP and published to Power BI Service on a scheduled refresh.
    • Replaced a manual spreadsheet export with a direct GraphQL API integration in Power Query — with pagination and a schema contract that halts the load if the source changes — now refreshing automatically twice a day over 1,100+ tickets.
    • Built a cash-flow dashboard unifying payables, receivables and bank balances, with recurring-expense projections for 36 suppliers under rules validated with finance.
    • Developed a reorder point and safety stock model for 2,100+ supplies, flagging stockout risk and telling management where the method breaks down.
    • Cut a semantic model's size by 46% with no metric changes, and fixed a circular Power Query evaluation that took refresh from 15+ minutes to 10 seconds.

    Power BI · DAX · Power Query · GraphQL · TOTVS Protheus

Coamo Agroindustrial

Campo Mourão, BR

Largest agribusiness cooperative in Paraná · data team of 60+ analysts

  1. Data Analyst

    Nov 2025 — Apr 2026

    • Built 5 Tableau dashboards from scratch for plant managers across 11 industrial plants, covering production, performance and product quality.
    • Maintained and evolved 20+ operational KPI dashboards, including migrating reports from Power BI to Tableau.
    • Wrote and optimized SQL and PL/SQL in Oracle for large-scale extraction, with dimensional modeling and ETL workflows.

    Tableau · SQL · PL/SQL · Oracle · ETL

  2. BI Intern

    Mar — Nov 2025

    • Hands-on training in BI, ETL, dimensional modeling and SQL over industrial datasets. Promoted to Data Analyst after 8 months.
Education

UTFPR

Federal University of Technology – Paraná

  1. Postgraduate Specialization in Applied Generative AI

    2026 — 2028 · in progress · Curitiba

    360 hours covering LLMs, RAG, autonomous agents, LLMOps and AI governance, focused on automation applied to data processes.

  2. BSc in Computer Science

    2022 — 2025 · Campo Mourão

    Thesis: deep learning for fault location in transmission lines — LSTM, GRU and Mamba compared over 168,000 simulated events (best: LSTM, mean error 0.90 km). The first work in the literature to evaluate Mamba in this domain.

Toolkit

BI & visualization

Power BI · DAX · Power Query (M) · Power BI Service · Tableau · Excel

Modeling

Dimensional modeling · Star schema · Semantic model optimization · Technical documentation

Data & integration

SQL · PL/SQL · ETL · REST & GraphQL APIs · TOTVS Protheus · dbt · BigQuery

Code & databases

Python (Pandas, NumPy) · Git · GitHub Actions · Oracle · SQL Server · DuckDB

Get in touch caiogoia123@gmail.com — or LinkedIn, whichever you prefer.
GitHub LinkedIn Résumé · EN Currículo · PT