Fabio – Python, LLM, RAG

An ongoing enterprise ERP modernization initiative migrating a production legacy .NET Framework and MySQL platform to a cloud-native stack using .NET 9, React Native, and Python/FastAPI. The project applies Specification-Driven Development and AI-agent orchestration via LangGraph, with autonomous coding agents executing implementation, refactoring, testing, and migration tasks under architectural governance. It serves as the production validation ground for the AAOE methodology — combining large-scale system modernization with AI-first engineering practices on a live codebase.

• Design and maintain the end-to-end modernization architecture, defining Clean Architecture and DDD domain boundaries.
• Define the migration strategy from legacy .NET Framework components to modern .NET 9 services, maintaining production stability throughout.
• Create detailed technical specifications for autonomous coding agents to implement, refactor, test, and validate code.
• Orchestrate LangGraph-based AI agent workflows for implementation, refactoring, testing, and migration tasks.
• Apply Specification-Driven Development to progressively replace legacy components without disrupting live operations.
• Design authentication architecture, API contracts, and service boundaries for the modernized platform.
• Apply and refine AAOE engineering principles across a real-world enterprise ERP modernization context.

A high-performance sports analytics platform combining computer vision, machine learning, and trajectory prediction to extract frame-level athletic performance insights from video and statistical data. The system processes video frame-by-frame and runs AI/ML models for trajectory analysis and performance metrics, designed for low-latency deployment on mobile and edge environments with a focus on computationally efficient inference.

• Design the end-to-end AI analytics architecture, defining data-processing pipelines and model-integration boundaries.
• Design and build computer vision pipelines for frame-level video analysis of athletic movement.
• Design and validate AI/ML models for trajectory prediction and athletic performance analytics.
• Define low-latency processing architectures for mobile and edge deployments.
• Design performance and memory optimization strategies for computationally intensive AI inference workloads.

An open engineering initiative formalising a methodology for combining AI agent orchestration with specification-driven development. AAOE coordinates a conversational model, a specification agent, and a coding agent so that business intent is encoded as machine-readable specs before implementation begins — eliminating ad-hoc prompting and making the entire development loop auditable. The methodology was field-tested across enterprise client engagements and published as open engineering standards.

• Design and maintain the three-agent AAOE architecture (prompt drafter, specification agent, coding agent).
• Apply and refine the methodology across production enterprise client projects.
• Author and publish the methodology standards as an open engineering initiative.
• Develop Python/LangGraph tooling and FastAPI services to support the agentic workflow.

An enterprise Agentic CRM platform for personalised financial-product campaigns driven by real-time customer interactions across three signal streams: AI chat and WhatsApp conversations, real-time web/app behavioural analytics, and customer data from a major Salesforce Personalization deployment. LangGraph orchestrated the multi-agent workflows; MCP handled Campaign API integration; and Pinecone-based RAG provided per-customer conversational history so agents would not re-offer declined campaigns to the same client. Designed to meet strict enterprise banking security and compliance requirements.

• Designed the full agentic CRM architecture and LangGraph multi-agent orchestration workflows.
• Built channel-aware routing for AI chat, WhatsApp, and real-time web/app interactions.
• Integrated Campaign APIs through MCP tool bindings.
• Designed RAG/vector retrieval via Pinecone to surface per-customer campaign history.
• Separated LLM-based reasoning from deterministic campaign eligibility and financial business rules.
• Designed secure integration boundaries between AI agents and internal enterprise banking APIs.
• Applied AI-agent orchestration principles to a production-oriented enterprise banking environment.

A unified enterprise CRM SDK standardising Salesforce CRM integrations, push notifications, and in-app messaging across mobile banking applications. Delivered across iOS (Swift/SwiftUI), Android (Kotlin), and backend services (Java/Spring Boot and Node.js/Express BFFs), establishing reusable SDK components and API contracts consumed by multiple mobile teams at a major Brazilian bank. AI coding agents were used throughout to accelerate specification, refactoring, and documentation — an early application of what later became the formal AAOE methodology.

• Designed the end-to-end CRM SDK architecture and technical roadmap.
• Built reusable SDK components for iOS and Android banking applications.
• Developed Java/Spring Boot backend services and a Node.js/Express BFF layer.
• Designed API contracts and integration boundaries with Salesforce CRM.
• Used AI coding agents to accelerate implementation, refactoring, testing, and technical documentation.
• Led backend, Android, and iOS engineers; performed architecture and code reviews.
• Worked directly with Product Owners to translate business requirements into technical solutions.