Diogo Martins de Assis.

• Built Open Finance data requests (investments, credit, FX) on a platform with 1M+ organic consents, adding 10 new data types.
• Work on Pix, bank slip and transfer flows handling 510k+ transactions/month — 78% of Limpa Nome deals use Pix.
• Improved idempotency of SQS/SNS payment events with Redis, cutting duplicate payment incidents by 40%.
• Open Finance payment initiation so users pay debts from other banks without leaving the app.
• Improved Datadog monitoring for Pix and bank slip flows, cutting incident detection to under 10 minutes.

C# · .NET · AWS · Kubernetes · Redis · GitLab CI/CD · Datadog · Databricks

• Key contributor to migrating a .NET WinForms desktop app to Angular + .NET Core, moving 60%+ of active users to the web in 18 months.
• Core contributor to an AI fiscal agent — an MCP server (MCP C# SDK) that lets Claude, ChatGPT and Cursor query fiscal data. Adopted by 2,000+ companies in 6 months.
• Co-built LLM invoice classification (CFOP, CST, NCM) with Semantic Kernel and OpenAI, reaching 92% accuracy.
• Co-developed the SEFAZ × ERP × SPED reconciliation engine, cutting reconciliation from days to under 2 hours.
• Contributed to the RabbitMQ/AWS pipeline capturing 270M+ documents/month, cutting latency by 35%.

MCP C# SDK · Semantic Kernel · OpenAI & Anthropic SDKs · RAG · C# · Angular · TypeScript · RabbitMQ · AWS · SQL Server

• Trained machine learning models on biometric eye-tracking data from subjects viewing architectural images, supporting a PhD thesis on neuro-affective architectural modeling that connects neuroscience and affective computing with parametric design.
• Contributed to the manuscript derived from the thesis, being prepared for submission to Frontiers of Architectural Research.

Machine Learning · Affective Computing · Neuroarchitecture · Python

• Investigating how instance selection (removing redundant and noisy training data) affects the efficiency and effectiveness of a retrieval-and-fusion system (RAG-Fuse) for extreme multi-label text classification (XMTC), with emphasis on rare ("tail") label performance.

Information Retrieval · RAG · XMTC

• Analyzing why model effectiveness often survives aggressive data reduction, using rigorous statistical testing to tell genuine data redundancy apart from insufficient statistical power.

Statistical Analysis · Python