Alessio Serra

I'm

About me in O(1)

I'm an AI researcher and engineer focused on multilingual large language models.

Thanks to our machine translation systems, this site is available in 32 languages. Hopefully, one of them is yours..

AI Engineer & Researcher

At Translated, I helped build Lara from scratch, a machine translation LLM that powers content for Airbnb, Uber, Shopify, and others, reaching over 200 million users.

  • Birth Place: Naples, Italy
  • Email: alessio.ser29@gmail.com
  • City: Rome, Italy
  • M.S. Degree: AI & Data Engineering
  • B.S. Degree: Computer Engineering
  • University: University of Pisa
  • Languages: Italian, English
  • Focus: Multilingual LLMs

Currently, I'm working on scaling Lara's training across more than 1,000 GPUs on CINECA's HPC cluster, with the goal of reaching the quality of the top 1% of professional translators.

Before that, I led a research initiative to expand language support in ModernMT, increasing production coverage from 56 to 201 languages.

I also co-founded Picarta.ai, a startup focused on using AI for image geolocalization.

What motivates me most is curiosity and the drive to turn ideas into something real. I'm always trying to think more deeply, grow through challenges, and learn from the people around me. Training is a constant in my life—whether it's fine-tuning LLMs, lifting weights at the gym, or preparing for a marathon.

Technologies & Tools

PyTorch

Deep Learning

Python

Primary Language

Slurm

HPC Job Scheduling

vLLM

High-Performance LLM Serving

Transformers

LLMs & NLP

MongoDB

Data Storage

Docker

Containerization

Git

Version Control

Current Focus

What I'm building and researching right now

Multilingual AI in Production

Building robust translation systems that handle real-world complexity at scale. Tackling challenges like code-switching, domain adaptation, noisy user generated content and maintaining quality across 200+ million daily users.

Scaling machine translation to 201 Languages

Expanding machine translation coverage while maintaining quality. Working with massive multilingual datasets and optimizing for low-resource languages using advanced transfer learning techniques.

Resume

AI Engineer and Researcher with extensive experience in building and scaling multilingual machine translation systems. Proven track record of serving 200+ million users through innovative LLM solutions.

Professional Experience

AI Engineer & Researcher

2022 - Present

Translated, Rome, Italy

  • Lara: Contributed across the full R&D pipeline of a large language model optimized for machine translation, including data collection, model training, alignment, and inference optimization. Developed within a startup-style team of four inside the company. Now the flagship B2B product, powering translation for Airbnb, Uber, Shopify, Nike, and others — reaching over 200M users globally. A B2C version was recently launched. Try it here.
  • Lara Grande: Played a key role in scaling the model to match the quality of top 1% professional translators. Led large-scale distributed training using over 1,000 GPUs on CINECA's HPC cluster.
  • Language Expansion: Initiated and led a research project expanding production coverage from 56 to 201 languages — a 4× increase — making it the first commercial MT engine to support this range. Defined the direction, implementation plan, and saw it through to delivery within eight months.
  • Instruction-following machine translation: Led research aligning LLM to follow detailed style guides using SFT and DPO.
  • Trust Attention: Proposed and validated a novel technique prioritizing high-value training data, achieving the most significant machine translation quality improvements in five years.
  • Polyglot: Developed a Language Identification model supporting 201 languages.

Awards & Recognition

Top 3% University Performer

2022

GetCredible

Named a top university performer by GetCredible, in recognition of exceptional talent and academic excellence among university students worldwide.

2nd Place - Loop Q Prize AI Competition

2022

Europe & Africa Region

Achieved second place in this international Cognitive Computing and Machine Learning competition. Built an emotion detection model for speech recordings that outperformed many competitors in both accuracy and innovation.

View on GitHub

Startup Experience

Co-Founder & CTO

2022 - 2023

Picarta AI, Italy

  • Co-founded AI startup as part of three-person founding team building image geolocalization platform
  • Developed the core model for image geolocalization using vision transformers and retrieval-based techniques
  • Gained valuable experience in fast-paced startup environment and product development
  • Learned crucial lessons about team dynamics, uncertainty management, and rapid iteration

Education

M.S. Artificial Intelligence & Data Engineering

2020 - 2022

University of Pisa, Pisa, Italy

110/110 summa cum laude (4.0 GPA). Specialized in Data Mining, Machine Learning, Computer Vision, Natural Language Processing, Optimization Theory, and Process Mining. Gained hands-on experience with Distributed Systems, Cloud Computing, and tools like MongoDB, Neo4j, Docker, Kubernetes, TensorFlow, and PyTorch.

B.S. Computer Engineering

2017 - 2020

University of Pisa, Pisa, Italy

110/110 (4.0 GPA). Strong foundation in computer engineering including mathematics, physics, algorithms, databases, computer architecture, computer networks, operating systems, and programming in C, C++, Java, Python, Matlab, SQL, JavaScript, and PHP.

Publications

Most of my research was not focused on publishing papers, but on building something people actually want. Rather than appearing in journals, the outcomes of my work are embedded in products that have seen real-world adoption at scale. For me, that's the most satisfying achievement.

Image Sentiment via Cross-modal Distillation

ECAI 2023 - Oral presentation
  • Designed a scalable cross-modal distillation framework leveraging textual sentiment to supervise visual sentiment classifiers using weakly-labeled social media data.
  • Built and curated a large-scale dataset of ~1.5M Twitter images with automated sentiment labels, eliminating reliance on manual annotation.
  • Achieved state-of-the-art in 5 benchmarks for visual sentiment analysis.

Paper website

Portfolio

A collection of my AI and machine learning projects, from academic research to production systems serving millions of users worldwide.

  • All
  • Machine Learning
  • NLP
  • Computer Vision
  • Data Engineering

NumpyGPT

A from-scratch GPT built with NumPy and Python's standard library. No autograd, no frameworks: every layer is re-implemented with its own forward and backward pass. Gradients are computed manually, updates are transparent, and every operation is spelled out.

T4SA-2.0

Cross-modal visual sentiment analysis system using knowledge distillation from text to vision models. Published at ECAI 2023, achieving state-of-the-art results on visual sentiment prediction.

Fake News Detection

Multi-modal AI system for misinformation detection using transformer architectures. Features stance analysis and source credibility scoring.

UniMusic

Scalable music discovery platform with hybrid MongoDB/Neo4j architecture. Handles millions of tracks with real-time recommendation algorithms.

Voice Vibes

Speech emotion recognition system with six novel AI architectures. Achieved 94% accuracy using ensemble methods and advanced feature engineering.

Federated DBscan

Privacy-preserving distributed clustering with federated learning. Enables collaborative ML without data sharing, achieving 99% privacy preservation.

PageRank Implementation

High-performance PageRank algorithm implementation with optimized sparse matrix operations. Handles graphs with millions of nodes efficiently.

Contact

Let's connect! Whether you want to discuss AI, machine learning, or potential collaborations, I'm always open to interesting conversations.

Location

Rome, Italy

Loading
Your message has been sent. Thank you!