Skip to content
Lorenzo Ciarpaglini

Lorenzo Ciarpaglini

Ph.D. Candidate in Artificial Intelligence·Sapienza University of Rome

I take things apart to see how they work, then explain them to whoever listens.

Education

View all(3)

Experience

View all(5)
Jan 2025 — Jan 2026

AI Engineer (part-time) · Oversell AI

Rome, Italy

Sole backend engineer of the LLM pipeline behind a sales-intelligence platform that analyzes hundreds of sales calls a day.

  • LangChain
  • Whisper
  • Redis
  • AWS
Sep 2024 — Dec 2024

Software Engineer (contract) · Hiresales

Rome, Italy

Backend of a vendor-hiring marketplace, built as one of two engineers and grown to ~1,000 active users.

  • Node.js
  • Express
  • MySQL
Jan 2022 — Mar 2023

Co-founder & Lead Engineer · AncientSociety

Rome, Italy

Built a browser-based game end to end and led a team of 6 engineers: 5,000+ players and ~$700k in revenue in its first 8 months.

  • React
  • Node.js
  • Solidity
  • Team Lead

Projects

View all(19)

Memory-Efficient Training with Block Coordinate Descent

Ph.D. research on block coordinate descent methods, one with a selective block rule, for training under tight GPU memory: only the active block keeps gradients and optimizer state, so memory drops to 2M + 16M/D bytes (M parameters, D blocks) — up to 95% less optimizer memory than Adam/AdamW. Benchmarked in fine-tuning and post-training (SFT, DPO) of language and vision models such as RoBERTa-Large and ViT-H, and extended to multi-GPU training with PyTorch FSDP. Ongoing.

  • PyTorch
  • FSDP
  • SFT
  • DPO
  • Optimization

Ground States of Disordered Spin Models

Recasts the 3D random-field XY model as a quadratic problem on a product of unit circles, so Riemannian local solvers apply, and drives them with Basin Hopping moves tailored to the physics. Finds lower-energy configurations than parallel tempering, simulated annealing and differential evolution within the same budget. Joint work with the Physics Dept. at Sapienza.

  • Python
  • Riemannian Optimization
  • Global Optimization
  • Statistical Physics

Enhanced Armijo Line Search for Deep Learning

A stochastic line search that reads the step size off the loss instead of a schedule: when an epoch fails to improve, it searches along the direction the epoch travelled, on a random sample of the training set. Beats Adam on validation loss in long regression runs and, with Adam, is the only optimizer tested that trains a 62M-parameter diffusion U-Net to recognisable digits. Code behind my M.Sc. thesis.

  • PyTorch
  • Optimization
  • Line Search
  • Diffusion Models

Satellite Collision Risk Prediction

Predicts the final collision risk of an orbital conjunction at least two days before closest approach, from Conjunction Data Messages alone. A cascade of 57 SVMs with MLP refiners scores 0.559 on the ESA Collision Avoidance Challenge test set — 2nd of 97 teams and 1st among fully ML-based entries. Team of 3; extended to satellite–debris conjunctions with the Italian Space Operations Command.

  • scikit-learn
  • PyTorch
  • SVM
  • Space

Teaching

View all(4)
2024/25 – 2025/26

Operations Research — Laboratory

Teaching Assistant · Sapienza University of Rome

Teaching the lab on integer and linear programming models in Gurobi — knapsack, bin packing, cutting stock with column generation, TSP with lazy constraints. With the Optimization and Decision Science lab, about 500 students; all materials are open source.

2025/26

Optimization and Decision Science — Laboratory

Teaching Assistant · Sapienza University of Rome

Teaching the course's optimization lab sessions.

2024/25 – 2025/26

Mathematical Programming

Guest Lecturer · Sapienza University of Rome

Two guest lectures on Transformer architectures.

Open Source

View all(1)

Centralized encoder freezing in PreTrainedPolicy, so layer freezing behaves the same in training and evaluation (under review).

  • Python
  • PyTorch
  • Robot Learning