Diogo Guimarães Silva

Full-stack developer · ML systems — Lisbon, Portugal

Three years shipping production web applications, now focused on data and ML systems: not just models, but the engineering that makes them trustworthy in production.

Selected work

energia-forecast dashboard: tomorrow's demand forecast, model vs baselines

energia-forecast

Live

2026 · personal project · in production

A day-ahead forecasting system for Portuguese electricity demand and MIBEL market prices, running autonomously on free-tier infrastructure. Four public data sources ingested daily, leakage-free rolling-origin evaluation, conformally calibrated intervals, and every forecast stored immutably and scored in public against outcomes.

2.8%demand MAPE (71-day backtest)
13.2 €/MWhprice P50 MAE vs 16.0 baseline
€0/moinfrastructure cost

Built end to end with AI-assisted development (Claude Code), with me as architect and reviewer; design decisions recorded as ADRs in the repository.

PythonLightGBMPostgreSQLFastAPI GitHub ActionsStreamlit

dr-watch

Live

2026 · personal project · in production

What did the Portuguese state decide today? An autonomous LLM system over the official gazette (Diário da República): every morning it ingests the day's acts, classifies them by theme, summarizes each in plain language with citations to the official PDF, and publishes a digest whose accuracy is measured continuously — number-grounding checks, citation validity, coverage — on a public accuracy page.

Same design DNA as energia-forecast, applied to LLM engineering: evals designed before prompts, immutable outputs, free-tier strict.

PythonGeminiLLM evalsNext.js PostgreSQLVercel

Experience

Software Developer — Bitcliq

September 2023 — present · Caldas da Rainha

Features and full applications across 7+ projects: Flask and Node.js APIs, React and C# frontends, React Native, and production debugging for external clients.

AI Development Intern — Banco BPI (CEIA)

August 2022 — July 2023 · Lisbon

OCR and document-processing tools for financial documents with Python, Tesseract, OpenCV and TensorFlow.

Full CV →

Writing

I write up what I build. First piece: The model was the easy part — notes from building energia-forecast, from backtests that lie by default to the scheduler that failed in production. All writing →