Tony INCORVAIA

• Built and delivered a lead-generation AI Agent across 65 school websites (110 campus and more than 300 000 students) increasing prospects conversion • Engineered the group's first knowledge base from scratch, designing robust data ingestion pipelines (web scraping, PDF extraction) and merge logic to unify disparate sources into a single source of truth. • Built the LLM evaluation harness and owned the end-to-end Azure deployment

– Designed and deployed a multi-tenant GenAI SaaS platform for creators. – Architected multi-provider LLM system routing across OpenAI, Anthropic and xAI with automatic provider selection and per-request cost tracking (37% cost reduction) – Built production RAG pipeline: Sentence Transformers embeddings → FAISS vector search → dynamic context injection with source attribution – Implemented MCP server (Model Context Protocol) enabling Claude Desktop and Cursor to query the knowledge base via Streamable HTTP transport – Developed real-time voice cloning and fine-tuned TTS model with streaming synthesis via sentence-chunked SSE – Designed complete multi-tenant architecture: per-tenant personas, knowledge bases, LLM configs, TTS engines – single codebase for multiple B2B clients – Built subscription billing with Stripe: checkout, in-place plan changes with SCA/3D Secure, proration, webhook sync, GDPR-compliant solution. Stack: Python, FastAPI, PostgreSQL, TypeScript, SvelteKit, FAISS, Sentence Transformers, FastMCP, Stripe

• Evaluated enterprise GenAI adoption strategies through technology watch and industry events (Nvidia, CAC40 partners), producing technical assessments of LLM implementation patterns and ROI for stakeholders. • Design & Developed a full-stack application reducing costs by 15%

– Within the Research & Development (R&D) : Python automation tool via CMake reverse-engineering for C++ devs Optimized SQL queries & C++ code SAFe Agile collaboration

– Built a full-stack application, reducing daily batch-checking time – Developed and delivered numerous Jira tickets for application maintenance and feature enhancements, ensuring 97% on-time delivery and high-quality code – Drove data analysis with Python (Pandas/NumPy): extracted insights from 2M + batch logs → identified 3 root causes of delays

– Built a working Substrate pallet in Rust for decentralized identity (KYC PoC) – Tweaked BABE consensus params – Connected a React + Polkadot.js frontend: users could mint & verify credentials