cv_sveva_pepe.txt

Sveva Pepe — AI/ML Engineer

Sveva Pepe is an AI/ML Engineer at Almawave Labs, Rome. Since 2025 she designs the AI architecture of an enterprise product built on LLM agents. Before that, from 2022, she extracted entities and relations from judicial documents for the public sector.

relations

  1. (Sveva Pepe, works_at, Almawave Labs)
  2. (Sveva Pepe, role, AI/ML Engineer)
  3. (Sveva Pepe, focus, AI architecture · LLM agents)
  4. (Sveva Pepe, worked_on, NER & relation extraction · legal text)
  5. (Sveva Pepe, published_at, AAAI 2022 · IEEE MLSP 2021)
  6. (Sveva Pepe, open_to, on-site roles in Europe)

Annotated by hand, in the format of the entity- and relation-extraction systems I build.

§1

Now — Almawave Labs, 2025 – present

Designing the AI architecture of an enterprise product

I'm responsible for the AI architecture: how LLM agents, workflows and services work together to process documents at scale. Follow a document through it:

Large-scale document-processing pipelines turn incoming documents into structured content the rest of the module can work on.

  • Python
  • Docling
  • PyMuPDF

An agent that writes the tests

A LangGraph agent that generates tests automatically (pytest, httpx) and is integrated into the team's pre-deployment testing framework.

  • LangGraph
  • pytest
  • httpx
  • Temporal WorkflowEnvironment
  • testcontainers

Owner of the AI architecture

I coordinate the technical work on the AI side and the key decisions on the AI architecture, and shape the product roadmap together with product management.

§2

Before — Almawave, 2022 – 2025

NLP and LLMs for judicial and administrative documents

Client delivery for public-administration clients: custom solutions on legal text, where language is long, formal and precise.

  1. Extraction

    LLM-based solutions for named entity recognition and relation extraction on judicial documents, including fine-tuning transformer models.

    • NER
    • Relation extraction
    • BERT
    • BART
  2. Prompting

    Prompt-engineering strategies designed for legal terminology and complex sentence structures.

  3. Evaluation

    Custom evaluation metrics and standardized preprocessing workflows, reused to improve efficiency across multiple government projects.

§3

Experience & education

  1. 2022 – present

    AI/ML Engineer — Almawave Labs

    Product development (2025 – present) · Client delivery (2022 – 2025). Almaviva Group; the company was renamed from Almawave to Almawave Labs in August 2026.

  2. 2021 – 2022

    Performance Engineer — Moviri, Milan

    Optimized performance-testing frameworks with LoadRunner, AppDynamics and Instana.

  3. 2020 – 2021

    Research scholarship — Sapienza University of Rome

    L3DAS project: built a 3D audio dataset and challenge for speech enhancement and sound event localization.

  4. 2019 – 2021

    M.Sc. Artificial Intelligence & Robotics — Sapienza

    110/110 cum laude, focus on machine learning and NLP. Certificate of Excellent Student, Sapienza, 2023.

  5. 2016 – 2019

    B.Sc. Computer Science & Automation Engineering — Sapienza

    110/110 cum laude.

§4

Skills

AI & LLM
LangChain, LangGraph, Pydantic AI, agent design (ReAct, multi-step reasoning, tool calling), RAG, prompt engineering, NER, relation extraction, text classification, Azure OpenAI
Backend & orchestration
Python, FastAPI, Temporal.io, Docker, Kubernetes (AKS)
Data & documents
MongoDB, Docling, PyMuPDF
Cloud & DevOps
Azure DevOps, CI/CD, Git, pytest, Locust

§5

Research

  1. AAAI 2022

    STEPS: Semantic Typing of Event Processes with a Sequence-to-Sequence Approach

    Sveva Pepe, Edoardo Barba, Rexhina Blloshmi, Roberto Navigli

  2. IEEE MLSP 2021

    L3DAS21 Challenge: Machine Learning for 3D Audio Signal Processing

    E. Guizzo, R. F. Gramaccioni, S. Jamili, C. Marinoni, E. Massaro, C. Medaglia, G. Nachira, L. Nucciarelli, L. Paglialunga, M. Pennese, S. Pepe, E. Rocchi, A. Uncini, D. Comminiello

Earlier university projects (NER, semantic role labeling, quaternion networks for 3D audio, reinforcement learning, robotics) are on GitHub.

§6

Contact

sveva.pepe@gmail.com

Based in Rome, open to relocation for on-site roles in Europe.