Alessandro Osti.
Senior Backend Engineer with over 25 years of experience. Enterprise systems, healthcare, telecommunications, banking, algorithmic trading, and AI engineering. Correggio, Italy.
I have been writing code since 1999. The first 19 years were spent on very different projects: healthcare management systems with complex Oracle architectures and PL/SQL optimization, native Android applications with NFC, GPS, and IoT device integration, telecom infrastructure with IVR systems and C++ drivers for contact centers. I worked as a consultant specializing in system integration, developing ERP middleware and migration tools for enterprise platforms.
Environments where software ships through rigorous processes and where a bug has real costs. That is where I learned to build robust software — not just software that works.
Since 2018 I am a Senior Backend Engineer at Lynx Group, working on enterprise systems for major Italian banking and insurance groups: developing REST and SOAP services on Java EE architectures with JBoss and Spring, integrated with clients' proprietary frameworks. Designing complex relational schemas and PL/SQL procedures on Oracle for critical business logic. I have worked on core banking platforms, on a retail mobile banking application for a major national bank, and on digital certification systems.
In parallel, in 2012, I started building MAOTrade, an automated trading system. What started as a project with my father became my permanent lab: Python, FastAPI, Docker, microservices — everything battle-tested in production, every day, for over 12 years.
Today I'm integrating AI engineering into my work — not as a pivot, but as a natural evolution of my method. ARIA is the proof: a financial analysis assistant built on multi-LLM architecture, RAG and streaming. I study, build, and document everything publicly. I also hold an MBA in Finance — accounting, management control, business planning, company valuation. A background that gave me a concrete understanding of how businesses work from the inside: skills I consider essential in any professional context, and that I apply to financial market analysis.
Over 20 years on mission-critical systems: banking, insurance, healthcare, telecom. REST and SOAP service development on proprietary frameworks, integration with legacy systems and ERP middleware, releases in regulated environments.
Integrating LLM models into existing systems. Multi-provider architectures, RAG with vector databases, semantic caching, SSE streaming. No prototypes — production systems with real data.
Design of complex relational schemas, advanced queries and PL/SQL procedures for banking and healthcare business logic. Financial time series, analytics, high-concurrency data management.
Containerization, custom orchestration, centralized logging, monitoring. Management of dedicated servers with 24/7 production services. Python and FastAPI for modern backends and high-performance APIs.
MAOTradeAutomated trading system in production for over 12 years. Microservice architecture with isolated Docker containers per broker account, automatic order management, real-time monitoring, backtesting, and cycle analysis.
ARIAFinancial analysis assistant built on top of MAOTrade. Multi-provider LLM architecture with Gemini, Claude, and GPT, RAG with Qdrant for querying market data, semantic caching, SSE streaming. Cycle analysis with channels, pattern recognition, fair value calculation.
RAG system for querying financial documents in natural language. PDF processing pipeline, local embeddings with SentenceTransformers, semantic search on FAISS, multi-LLM integration (Claude, GPT). REST API with FastAPI and a Streamlit demo interface.
The experience of building ARIA, a financial analysis system in production, using Claude Code for the entire development — backend, frontend, deployment. An honest account of how it went: the five-phase method that worked, the sessions that failed, and one thesis: AI coding tools amplify what you already have.
Structured AI engineering study path documented publicly: LLM architecture, tokenization, RAG, AI agents, Model Context Protocol. Includes hands-on vibe coding experiments and reflections on learning methodology.