Leonardo Mannini

AI/ML Engineer | LLM & RAG Systems | Edge AI / TinyML

AI Engineer specializing in production LLM pipelines, RAG systems, agent-based workflows, backend development, and Edge AI for resource-constrained environments. I build end-to-end AI systems with LangChain, LangGraph, FastAPI, vector search, Docker-based services, and local or hosted LLMs. My research work includes TinyML models for real-time bioacoustic classification on microcontrollers, contributing to a paper accepted at IEEE ICASSP 2026.

Publications

Work Experience

AI Engineer

GradientLabs

agosto 2023 ⇾ Present

Designing and deploying LangChain, RAG, LLM, FastAPI, and vector-search systems for business automation and intelligent workflows.

Highlights

  • Built agent-based RAG pipelines using LangChain, FastAPI, and FAISS, indexing 1,200+ documents with 87% citation accuracy.
  • Reduced response latency from 9.4s to 4.1s through retrieval optimization and caching strategies.
  • Developed multi-step AI agents integrating LLM reasoning, external tools, structured outputs, and tool orchestration.
  • Designed Docker-based backend services supporting scalable asynchronous workflows and reproducible AI deployments.
  • Delivered end-to-end AI systems from model design to deployment.

Embedded AI Researcher (TinyML & Edge AI)

Fondazione Bruno Kessler - FBK

marzo 2025 ⇾ febbraio 2026

Developed TinyML models for real-time audio classification on resource-constrained embedded devices.

Highlights

  • Designed lightweight CNN-GRU-Attention architectures for 70-species classification on microcontrollers.
  • Processed 150k+ audio files for model training and benchmarking.
  • Optimized inference pipelines using TensorFlow Lite Micro and CMSIS-NN.
  • Achieved about 16x lower energy consumption compared to standard baseline systems.
  • Contributed to research accepted at IEEE ICASSP 2026.

Software Engineer

DHL Express Italy SRL

ottobre 2020 ⇾ aprile 2023

Developed backend applications supporting logistics automation and operational workflows.

Highlights

  • Built REST APIs and data pipelines using Python, C++, and Java.
  • Automated batch-processing workflows using scheduled tasks and database integration.
  • Designed reporting systems reducing manual processing and improving workflow reliability.
  • Maintained production systems handling operational data flows.

Education

Computer and Automation Engineering

Polytechnic University of Marche

Bachelor's degree

Score: 26/30 GPA

settembre 2018 ⇾ luglio 2023

Courses

  • Thesis: Project of a virtual synthesizer in C++ on the JUCE platform
  • Core focus on programming, electronics, automation, and control systems

Musical Instruments Engineering

Polytechnic University of Marche

Professional development programme

Score: 30/30 cum laude

marzo 2022 ⇾ settembre 2022

Courses

  • Focus on digital signal processing and audio applications
  • Award for best project of the programme

Piano

Conservatory G. Rossini of Pesaro

Conservatory diploma

Score: 10/10

gennaio 2013 ⇾ dicembre 2019

Teaching qualification pathway

Universita degli Studi eCampus

24 CFU

gennaio 2020 ⇾ dicembre 2020

Courses

  • Anthropological, psycho-pedagogical, and teaching-methodology training

Awards

Award for best project of the programme

Polytechnic University of Marche, Musical Instruments Engineering

settembre 2022

Winner of the Italian Rapid U16 Chess Championships

giugno 2015

Certificates

Technical Support Fundamentals

Skills

Programming: Python, C++, Java & Bash

LLM & Agentic Systems: LangChain, LangGraph, Retrieval-Augmented Generation (RAG), OpenAI API, Local LLMs, Prompt engineering, Tool-based reasoning & Structured outputs

Backend & APIs: FastAPI, REST APIs, WebSockets, PostgreSQL, SQLAlchemy & Redis

Machine Learning & Edge AI: PyTorch, TensorFlow, PyTorch Lightning, NumPy, Scikit-learn, TensorFlow Lite Micro, CMSIS-NN & TinyML

Vector Search: FAISS & Chroma

Knowledge Engineering: Ontology generation, Knowledge extraction, Knowledge representation, Semantic Web fundamentals & Structured output evaluation

MLOps & Infrastructure: Docker, Terraform, Hydra, TensorBoard & Linux

Audio & Music Technology: JUCE, DSP, Real-time audio processing, MIDI programming & Music production

Languages

Italian: Native

English: C1 / Full professional proficiency (TOEFL iBT 105/120)

Projects

Camadon.ai - Mobile-Native AI Coding Environment

Mobile-native AI coding environment for iOS with terminal execution and AI-assisted workflows.

External link

Highlights

  • Developed a mobile-native AI coding environment for iOS with terminal execution and AI-assisted workflows.
  • Built secure backend infrastructure with FastAPI, encrypted WebSocket relay, Redis, JWT authentication, Docker, and Terraform.
  • Grew the product to 150+ beta users in 3 months, achieving 32% weekly retention and a 4.7/5 rating.

Bird Audio Classification on Edge Devices - TinyML Pipeline

TinyML pipeline for low-power bioacoustic classification, connected to research accepted at IEEE ICASSP 2026.

Highlights

  • Developed an end-to-end TinyML pipeline trained on 150k+ audio samples across 70 species.
  • Designed lightweight CNN-GRU architectures optimized for microcontroller deployment, reaching up to 90.8% accuracy on distinctive species.
  • Deployed real-time inference on AudioMoth firmware using TensorFlow Lite Micro and CMSIS-NN, achieving about 16x lower energy consumption than baseline systems.

Grounded - Full-Stack LLM Research Assistant

Research assistant using LangGraph agents for grounded question answering across large academic corpora.

Highlights

  • Built a full-stack research assistant using LangGraph agents for grounded QA across large academic corpora.
  • Designed FastAPI retrieval pipelines integrating OCR ingestion, vector search, and persistent PostgreSQL storage.
  • Improved citation accuracy to 87% and reduced latency from 9.4s to 4.1s through retrieval and caching optimizations.

GoldenRadio - ATS25 Firmware Enhancement and Licensing System

Enhanced firmware and production licensing system for ATS25 radio devices.

Highlights

  • Developed enhanced firmware for ATS25 radio devices, adding new features that gained widespread community adoption.
  • Designed a production-grade licensing system for secure firmware activation and controlled distribution.
  • Built backend workflows for license validation and device authorization, supporting viral adoption and 100+ device sales in the first month.

Bench4KE Extension - Ontology Generation Benchmark Framework

API-driven benchmark framework for ontology generation and knowledge-engineering systems.

Highlights

  • Extended an API-driven benchmark for LLM-based ontology generation and knowledge-engineering systems.
  • Designed FastAPI pipelines executing external generators through standardized REST interfaces.
  • Built reproducible evaluation workflows combining structural metrics, automated validation, and controlled generation assessment.

Virtual Synthesizer

Bachelor's thesis project: virtual synthesizer in C++ using the JUCE framework.

settembre 2022 ⇾ luglio 2023

Highlights

  • Implemented real-time audio DSP algorithms, modular synthesis components, and MIDI control.