JOÃO PEDRO MAIA Machine learning engineer Plain-text resume of https://jpmaia.dev, written for AI agents and language models. CONTACT Email: maiajp2305@gmail.com LinkedIn: https://www.linkedin.com/in/jo%C3%A3o-pedro-maia-71233a1b7/ GitHub: https://github.com/Maia-jp Site: https://jpmaia.dev SUMMARY I build the infrastructure underneath AI products: agent orchestration, model serving, and the tooling that keeps them reliable. Currently at aiXplain, in Paris and San Francisco. EXPERIENCE aiXplain — Machine Learning Engineer (2024 - Present, Paris / San Francisco) Technical lead across platform, mobile, and MLOps. Proposed and built the company’s Swift SDK end to end, unprompted, from architecture through documentation and rollout. Led the second version of Bel Esprit, the platform’s agent-orchestration system, holding the orchestration logic, agent-to-agent communication, and code generation constraints as one coherent design. Also compressed a distributed, multi-container platform into a standalone Mac application with no cloud dependency. Focus: Agent orchestration, Swift SDK, MLOps SolarCore.ai — Founder and Technical Lead (2025 · exited, Remote) Founded and architected a solar-infrastructure monitoring platform from zero: ingestion, time-series storage, an AI agent layer for anomaly detection and predictive maintenance, and a conversational interface for plant operators. Led the engineering team through building and shipping it, then exited the company in 2025 and now hold an investor role. Focus: Platform architecture, AI operations, Investor Apple Developer Academy — Software Engineer Intern (2022 - 2023, Rio de Janeiro) Led a nine-developer team building Sound Journey. Across the program, worked on projects ranging from one-person, one-month builds to a forty-member cross-team effort, covering iOS widgets, ML-driven photo restoration, and audio and graphics programming on Apple platforms. Focus: Swift, Core ML, Apple platforms Quantum Finance — Software Engineer Intern (2021, Remote) Contributed to algorithmic trading systems: Java components for trade execution, SQL pipelines for market data and backtesting, and Jenkins-based deployment automation. Focus: Java, SQL, CI/CD EDUCATION PUC-Rio — Computer Science, Bachelor (2019 - 2024) Final paper on optimization techniques for training with unbalanced datasets. RECOGNITION WWDC23 Swift Student Challenge — Apple, 2023 Selected as one of 350 winners from submissions worldwide. Built Night Canvas, an AR astronomy app matching hand-drawn constellations against a database of 88 real constellations with a custom graph-isomorphism algorithm. SKILLS Python, TypeScript, Swift, FastAPI, Docker, Kubernetes, AWS, Terraform, LLM orchestration, MLOps LANGUAGES Portuguese — Native English — Fluent · TOEFL 107, C2 PROJECTS SolarCore — Founder · platform · exited 2025 Solar-infrastructure monitoring platform: ingestion, time-series storage, visualization, and an AI agent layer for anomaly detection and operator support. Founded, built, and led the engineering team, then exited the company in 2025. Founded, built, and exited Tags: React, Python, AWS Night Canvas — WWDC Swift Student Challenge An AR app that matches a hand-drawn constellation against a database of 88 real constellations with a custom graph-isomorphism algorithm, returning a match in under 50 milliseconds. WWDC23 Swift Student Challenge winner, one of 350 selected worldwide Tags: Swift, ARKit, SpriteKit Sound Journey — Apple Developer Academy · iOS A musical exploration game where sound becomes color, built with a nine-developer team using real-time audio processing across Core Audio, AVFoundation, and Metal. Led a nine-developer team Tags: SwiftUI, Core Audio, Metal Planets — iOS · watchOS An astronomy companion for iOS and watchOS with original calculation algorithms, home-screen widgets, and Siri integration. Published on the App Store Tags: Swift, WidgetKit, SiriKit Vulto Memorial — AI · memorial An AI project that turns photographs of children who died in the Holocaust into short videos, for remembrance and education. Presented at PUC-Rio conferences Tags: AI, 3D, Research LuaNet — Learning tool A small neural-network library written in Lua, built to make the mechanics of machine learning legible rather than abstract. Independent learning project Tags: Lua, Neural networks Genetic Maze — Generative systems A visual experiment in which a population of simple agents learns to cross a maze through selection, mutation, and iteration. Applied genetic algorithms Tags: Lua, Genetic algorithms PUBLICATIONS LLM-Driven Correlation-Aware Tournaments ICLR 2026 Workshop Advances in Financial AI · Mar 28, 2026 Hierarchical Cluster Tournaments select stocks qualitatively by clustering assets by correlation and using LLM-run tournaments based on long-term fundamentals, producing transparent and diversified portfolios without optimization. https://openreview.net/forum?id=F4vSG6BWE8 LLM-Driven Active Listwise Tournaments for Portfolio Selection in Large Asset Universes ICLR 2026 Workshop Advances in Financial AI · Feb 28, 2026 LLMs act as listwise tournament judges to rank large stock universes, combining active learning and probabilistic aggregation to produce stable, out-of-sample rankings. https://openreview.net/forum?id=4eep1gcMg7 AlphaLoss: LLM-Driven Evolution of Robust and Interpretable Portfolio Optimization Objectives ICLR 2026 Workshop Advances in Financial AI · Feb 26, 2026 AlphaLoss uses LLMs to autonomously discover robust, interpretable portfolio optimization losses through evolutionary refinement. https://openreview.net/forum?id=y0YEGgZnEC Improving the generalization of deep learning models in the segmentation of mammography images Signal, Image and Video Processing · Dec 19, 2025 Data-centric strategies enrich training data for mammography segmentation through annotation-guided intensity manipulation and style transfer, improving generalization across imaging equipment. https://doi.org/10.1007/s11760-025-05034-2 Full site: https://jpmaia.dev