Riccardo Niccolò Agosti, Data Scientist & AI Engineer: data science and LLMs

Hi, I’m Riccardo Niccolò Agosti. I build LLM pipelines and machine learning models.

Data Scientist & AI Engineer at Leithà, Unipol Group.

FLORENCE, IT · 2026Get in touch
Core focus

“I build AI systems and test them like a statistician would.”

I take problems from raw data to evaluated models.

What I build

LLM pipelines and agents with LangChain, LangGraph and PydanticAI, machine and deep learning models in PyTorch and Scikit-Learn, and statistical analyses in R, SQL and Stata.

Data Scientist & AI Engineer at Leithà (Unipol Group), where I started as an intern.

Foundations

A BSc in Statistics from Florence and an MSc in Data Science from Padova: inference, machine learning, deep learning and optimization.

Riccardo

In numbers

Generative AI and LLM agents,

machine and deep learning,

and the statistics behind them.

Years of statistics & data science5
Market research interviews300+
English levelC1

A network that follows your pointer: move it, or drag to spin.

From statistics in Florence
to LLM pipelines at Leithà.

I studied Statistics in Florence, ran more than 300 market research interviews in retail stores for Modus, and graduated in Data Science from Padova. From February to June 2026 I was an intern at Leithà, part of the Unipol Group, building proofs of concept that read scanned documents with OCR, hand them to LLMs and return structured data, then measuring how well they do. Now I work there as a Data Scientist & AI Engineer. I like problems where the data is messy and the requirements change halfway through.

Tech stack & ecosystem

Python
R
SQL
Stata
PyTorch
TensorFlow
Keras
Scikit-Learn
XGBoost
ONNX
LangChain
LangGraph
PydanticAI
Google Cloud
AWS
Azure
Apache Spark
Tableau
Power BI
Git
Linux
Notion
Feature — 01

Generative AI systems

LLM pipelines and agents: multi-prompt, agentic and tool-use prompting, OCR for scanned documents, orchestration with LangChain, LangGraph and PydanticAI, and an evaluation for every pipeline.

LangChain
LangGraph
PydanticAI
Prompt engineering
OCR pipelines
LLM evaluation
Feature — 02

Machine & deep learning

Models for images, sequences and tables in PyTorch, TensorFlow, Scikit-Learn and XGBoost: object detection, time-series forecasting, classification and clustering.

PyTorch
TensorFlow
Scikit-Learn
XGBoost
Computer vision
Time series
Feature — 03

Statistics & optimization

Statistics came first: inference, field data collection for market research and continuous optimization, in R, SQL, Stata and Python.

Inferential statistics
R
SQL
Stata
Optimization
Market research

A few of the things I have built, with the code on GitHub

Scroll to explore

Projects

Selected work

A few examples of what I build. Many more projects are on my GitHub.

01Computer vision

Pokémon detection, three ways

YOLOv11s, RT-DETR and Faster R-CNN trained on the same nine Pokémon classes and compared on speed, classification and box quality.

per frame with YOLOv11s, about 8× faster than Faster R-CNN
11 ms
F1 for YOLOv11s, the best of the three
0.89
mAP 50-95 for Faster R-CNN, the tightest boxes
0.80
  • PyTorch
  • Ultralytics
  • YOLOv11s
  • RT-DETR
  • Faster R-CNN
  • OpenCV
Case study

02Time series

Weather, 30 steps ahead

A Seq2Seq GRU in PyTorch that reads 90 steps of history and forecasts 76 weather variables 30 steps ahead, for 422 stations. Built for the Kaggle challenge of a deep learning course.

weather variables forecast by one model
76
stations, 192,432 training sequences
422
best validation MAE over 30 steps
0.557
  • PyTorch
  • GRU
  • Seq2Seq
  • Scikit-Learn
  • Pandas
Case study

03Optimization

Max clique, made continuous

The maximum clique problem recast as continuous optimization over the simplex, and solved with projected gradient descent and three variants of Frank-Wolfe. Built for the Optimization for Data Science course in Padova.

execution time on the DIMACS graphs
−30%
found the largest cliques on average
PGD + L0
moved the least when the step size changed
FW, AFW
  • Python
  • NumPy
  • SciPy
  • Frank-Wolfe
  • Projected Gradient Descent
  • DIMACS
Case study
More on GitHub

Professional Journey

Roles and degrees in order, from a shop floor in Dublin to Data Scientist & AI Engineer at Leithà.

  1. Shop Assistant

    Liberty Recycling Training & Development. Dublin, Ireland. Jun — Jul 2019.

    A summer on a busy shop floor in Dublin: the till, the stock and the customers, all in English.

    • Point-of-sale operations and cash handling.
    • Inventory for a high-volume retail shop.

    EN: Two months working in English

  2. BSc Statistics

    Università degli Studi di Firenze. Florence, Italy. Sep 2021 — Jul 2024.

    Three years of probability, inference and applied statistics, taught in Italian.

    BSc: Statistics

  3. Market Data Collector

    Modus S.r.l.. Florence, Italy. Sep 2023 — Feb 2024.

    Field work for market research, alongside the last year of the BSc.

    • Ran more than 300 quantitative and qualitative interviews in retail stores.
    • Collected the primary data behind market research studies.

    300+: Interviews in retail stores

  4. MSc Data Science

    Università degli Studi di Padova. Padova, Italy. 2024 — Sep 2026.

    Machine learning, deep learning and optimization, taught in English.

    MSc: Data Science

  5. AI Data Science Intern

    Leithà S.r.l., Unipol Group. Milan, Italy. Feb — Jun 2026.

    Proofs of concept that turn scanned documents into structured data.

    • OCR first, then LLMs that extract structured data with multi-prompt, agentic and tool-use prompting.
    • Orchestrated workflows and agents with LangChain, LangGraph and PydanticAI, and measured each pipeline.
    • Ran ML and AI experiments on Google Cloud, AWS and Azure as the requirements changed.

  6. Data Scientist & AI Engineer

    Leithà S.r.l., Unipol Group. 2026 — Present.

    Back at Leithà after the internship, now as Data Scientist & AI Engineer.

Education & languages

Statistics first, then data science.

A bachelor’s in Statistics, a master’s in Data Science taught in English, and English at C1.

See the journey

Master of Science

September 2026

Data Science

Università degli Studi di Padova

Degree

MSc

Riccardo Niccolò Agosti

Taught in English

Official degree badge

Bachelor of Science

July 2024

Statistics

Università degli Studi di Firenze

Degree

BSc

Riccardo Niccolò Agosti

Taught in Italian

Languages

English level

Italian & English

Italian, native · English, C1

Level

C1

Riccardo Niccolò Agosti

Worked in English in Dublin, studied in English in Padova

Skills

Thetoolbox,cardbycard.

Grouped by what I use them for. Open a card to see everything in it.

Generative AI

LLMs and agents,
put to work in pipelines.

Main tools

  • LangChain
  • LangGraph
  • PydanticAI

OCR, then LLMs with multi-prompt, agentic and tool-use prompting, orchestrated and evaluated.

Machine & deep learning

Vision and sequences,
trained in PyTorch.

Main tools

  • PyTorch
  • Scikit-Learn
  • XGBoost

Object detection, sequence models and forecasting, and classic ML for classification and clustering.

Data, cloud & tools

From queries to dashboards,
and experiments in the cloud.

Main tools

  • Python
  • SQL
  • Tableau

Python, R, SQL and Stata for analysis; Spark, Tableau and Power BI at scale; Google Cloud, AWS and Azure for ML.

Tools I reach for
when the data won’t behave.

PythonPyTorchLangGraphSQLRStatisticsXGBoostOCRGoogle CloudStata
Data Scientist & AI EngineerLeithà · Unipol Group

Move over the box

Let’s talk about
LLM pipelines

Write to me in Italian or English, by email or through the contact page.

Contact Me
Email Me
Data ScienceAI EngineeringGenerative AILLM pipelinesMachine LearningStatisticsForecastingOptimizationData ScienceAI EngineeringGenerative AILLM pipelinesMachine LearningStatisticsForecastingOptimizationData ScienceAI EngineeringGenerative AILLM pipelinesMachine LearningStatisticsForecastingOptimizationData ScienceAI EngineeringGenerative AILLM pipelinesMachine LearningStatisticsForecastingOptimization