Luca Sargenti

• Build and maintain the shared design system: performant, Signal-based, accessible Angular components reused by every product team.
• Lead the refactoring of core LMS features central to the product, with Angular Signals, RxJS and state management at scale.
• Set up CI/CD pipelines and the microfrontend architecture; maintain and release the shared component libraries.
• Built MCP tooling for internal libraries; apply prompt and context engineering to cut Claude Code token usage in day-to-day development.

• Led the frontend of the web translation editor that replaced TRW's legacy software; the whole company now translates in it.
• Cut the time to open a large project from 5–10 minutes to near-instant with infinite and virtual scrolling over thousands of variable-height segments.
• Built a custom input engine on contenteditable: own caret, selection, clipboard and undo/redo, per-language typography and right-to-left text, with inline tags protected from edits.
• Kept it smooth at scale with row-level Angular Signals, an R-tree spatial index for spellcheck hit-testing, and server-side QA streamed back in real time.

• Designed and built a serverless backend from the ground up using AWS Lambda, DynamoDB, SQS, SES, S3, and IoT for smart device operations.
• Delivered the business web app and prepared full handoff documentation for team transition.

• Built most of the core product over four years, working across web, mobile, backend, infrastructure and embedded UI.
• Built the web platform from scratch in Angular and its mobile counterpart in React Native.
• Architected microservice-based REST APIs in Node.js with gRPC inter-service communication, RabbitMQ, and Redis.
• Ran the Docker / Docker Swarm infrastructure and the network layer (Faucet); built a touch-device UI with Qt on Sway.

A sixteen-module AI engineering curriculum I run on my own. Retrieval, agents, tool calling and MCP, built by hand in TypeScript, with every exercise ending in a measurement.

• Resilient multi-provider LLM client in TypeScript (Gemini, Ollama, OpenRouter): retries with jittered backoff, streaming, cancellation, cost estimates and routing; 129 automated checks.
• Hybrid search on PostgreSQL + pgvector fused with Reciprocal Rank Fusion, measured with recall@10 and MRR; agentic retrieval with Corrective RAG and Self-RAG.
• Cut local-model prefill from ~30 s to 0.3 s on a 6,200-token prompt by reordering it for prefix caching.

Stack: TypeScript, Python, LLMs, RAG, pgvector, Agents, Tool calling, MCP, Evals

luca-sargenti-portfolio.vercel.app/projects/ai-engineering-lab

A local-only expenses tracker. Drop in bank statements as PDF, CSV or XLSX, and a local LLM extracts, deduplicates and classifies every transaction into your own categories.

Stack: Python, FastAPI, SQLModel, SQLite, React, TanStack Query, Ollama, LLMs

luca-sargenti-portfolio.vercel.app/projects/personal-expenses-tracker · github.com/LucaSarge4/personal-expenses-tracker

A local-first AI coach that syncs workouts from Garmin and Strava, embeds session notes into pgvector, and answers training questions through a LangGraph agent running on a local Ollama model.

Stack: Python, LangGraph, FastAPI, PostgreSQL, pgvector, React, Ollama, Docker

luca-sargenti-portfolio.vercel.app/projects/ai-training-coach

A real-time auction room for a ten-person fantasy football league. Players bid from their phones, the auctioneer runs the room from a console, and a TV shows every squad live.

Stack: Next.js, React, TypeScript, Supabase, PostgreSQL, Realtime, Vitest

luca-sargenti-portfolio.vercel.app/projects/fantacalcio

A personal dashboard for ISINs. Paste certificates, stocks, ETFs and bonds, get official EU reference data, and price them with your own provider keys.

Stack: TypeScript, Next.js, React, Supabase, Vitest

luca-sargenti-portfolio.vercel.app/projects/isinboard · github.com/LucaSarge4/isinboard