AI / ML ENGINEERING / MSc AI @ UniCal

Hi, I'm
Paolo Pangallo.

I build reliable AI systems.

I'm completing an MSc in Computer Engineering, specializing in Artificial Intelligence. I work on LLMs, machine learning and software systems, connecting models, data and practical applications.

  • LLM & Agentic Systems
  • GraphRAG & Verification
  • Software Engineering
Explore featured work ↗Download CV (PDF) ↓Email me ↗

Featured work: MCP E-CommerceMTB GraphRAGCephalopod AI

01 / Selected work

Selected projects.
Behind the code.

Three projects, each connected to specific implementation details and verifiable source code.

ARCHITECTURE / 01AI APPLICATION
A shopping request becomes a tool-driven workflow.
⌕USER INTERFACEChat & Search
→
✳ORCHESTRATIONAI Agent
→
⌘INTEGRATIONSMCP Tools
EBAY APIRETRIEVALREDISPOSTGRESQL
Simplified architecture · explore actual code in the repository
01 / Featured engineering projectFull-stack + AI

MCP E-Commerce.

An AI-assisted shopping platform integrating agent orchestration, MCP tools, eBay services, retrieval and a React frontend.

IMPLEMENTATION / REQUEST TO APISource-linked walkthrough

Follow one implementation path: agent coordination, explicit MCP tool handlers, and a test for streaming API behavior.

  1. 01Agent orchestration

    Code coordinating an AI request

    Inspect source
  2. 02MCP tool handlers

    Code connecting the agent with application tools

    Inspect source
  3. 03API test

    Agent streaming-mode test

    Inspect source
Problem
Turn flexible product requests into explicit actions while keeping the application usable.
Engineering
Tool integration, API design, search and ranking, state management and persistence.
Stack
Python · FastAPI · MCP · React · PostgreSQL · Redis · Qdrant
Engineering notebook / 01Explore the choices

Architecture

Explicit agent and tool boundaries.

Rather than treating a shopping request as a single prompt, the application separates agent orchestration, MCP tools, search services and product-facing APIs.

Inspect related code
Technical notes
The repository separates agent orchestration, MCP tooling, search services and application APIs. External eBay/LLM integrations require configuration. The project also includes automated Python and frontend build checks.
Read code & documentation
RESEARCH / 02MASTER'S THESIS
Traceable evidence before generated conclusions.
CaseKGDocsClaimcheckTraceDossierReview
Clinical caseKnowledge + documentsClaim verificationEvidence + provenanceHuman review
GRAPHPROVENANCERETRIEVALVERIFICATION
Research system · conceptual representation, not clinical validation
02 / Research projectLLMs + GraphRAG

MTB GraphRAG.

My master's thesis explores an architecture for evidence-grounded AI support in Molecular Tumor Boards, with a focus on traceability and verification.

IMPLEMENTATION / EVIDENCE AND CHECKSSource-linked walkthrough

Inspect how retrieval and verification are implemented. These are source references, not claims of clinical performance.

  1. 01Evidence pipeline

    Retrieval and processing code

    Inspect source
  2. 02Runtime checks

    Tests exercising the agentic runtime

    Inspect source
  3. 03Architecture notes

    Documented system boundaries and design

    Inspect source
Problem
LLM-generated statements need inspectable sources, not just plausible language.
Engineering
Structured knowledge, retrieval, claim verification and evidence provenance.
Focus
GraphRAG · Knowledge Graphs · LLMs · Reproducibility
Engineering notebook / 02Explore the choices

Architecture

Inspect the evidence behind a claim.

Retrieval and verification are separate concerns. The architecture is designed around structured sources, provenance and explicit handling of unsupported statements.

Inspect related code
Research boundaries
This is a research prototype, not a medical device. It does not provide autonomous therapeutic decisions. Different experimental configurations may require distinct datasets, service endpoints and evaluation conditions.
Explore the architecture
EXPERIMENTS / 03CLASSICAL + LEARNED AI
Different agents. Different ways to search.
MINIMAXMCTSREINFORCEMENT LEARNINGNEURAL AGENTS
Conceptual illustration of search and learning approaches
03 / Machine learning & searchExperimental AI

Cephalopod AI.

A game-playing AI project investigating heuristic search, Monte Carlo Tree Search, reinforcement learning and neural approaches.

IMPLEMENTATION / INSPECT THE STRATEGIESSource-linked walkthrough

Review the actual game rules and strategy implementations, with design notes for comparing methods fairly.

  1. 01Game mechanics

    Board state and rules

    Inspect source
  2. 02Decision strategies

    Implementations of the explored agents

    Inspect source
  3. 03System design

    Architecture and assumptions

    Inspect source
Problem
Explore alternative strategies for selecting actions in a game environment.
Approaches
Minimax / Alpha-Beta, MCTS, behavioral cloning and AlphaZero-style agents.
Stack
Python · PyTorch · Optuna · Reinforcement Learning
Engineering notebook / 03Explore the choices

Architecture

Different decision-making strategies.

The repository explores classical search alongside learning-based agents. These approaches have different computational needs and evaluation assumptions.

Inspect related code
Experimental notes
The methods use different assumptions and training setups, so their outcomes should not be compared without matching evaluation conditions. The repository documents the variants and their limitations.
Examine the experiments
02 / System explorer

See how the
systems work.

A deeper look at the architectures, with clickable stages and direct links to the source code.

Explore featured projects ↗
system.explorer Interactive overview

Explore an architecture

01 / Input↗

User request

A natural-language shopping request starts the workflow.

View related source ↗
ARCHITECTURE / CONCEPTUALSELECT A NODE TO INSPECT
03 / Project directory

More work,
different problems.

Find a project by discipline or keyword. Open the repository to inspect the implementation, setup and limitations.

8 projects
Browse every GitHub repository ↗Source code and documentation are public.
04 / Engineering approach

From prototypes
to engineering choices.

My projects span research and software. The common thread is making system behavior understandable, testable and connected to real requirements.

01 / Integration
↗

Models are one component.

Applications also need APIs, data flows, external tools, interfaces and operational boundaries.

Example: MCP E-Commerce ↗
02 / Evidence
⌁

Outputs need context.

In retrieval-heavy systems, source traceability and explicit uncertainty are design requirements.

Example: MTB GraphRAG ↗
03 / Experimentation
◎

Comparisons need care.

Models, strategies and benchmarks are meaningful only when setup and limitations are clear.

Example: Cephalopod AI ↗
Technology map

What I use.

Technologies appearing across my projects. Not a claim of expert-level proficiency in every tool.

AI / ML

PyTorchscikit-learnLLMsRAGGraphRAGMCPReinforcement Learning

Backend / data

PythonFastAPIJavaSpring BootPostgreSQLRedisNeo4jQdrant

Frontend / tools

ReactTypeScriptAngularDockerGitLinuxREST APIs
05 / Background

A little more
about me.

I'm finishing a master's degree in Computer Engineering, specializing in Artificial Intelligence, at the University of Calabria.

I like turning ideas into working software, understanding the trade-offs, and documenting what a system can actually do.

Email me ↗LinkedIn ↗GitHub ↗
Ritratto di Paolo Pangallo
PERSON / PROFILE

Paolo Pangallo

Computer Engineering · AI

Rende / Calabria, Italy
2024 — current
M.Sc. Computer Engineering

Artificial Intelligence · University of Calabria · Thesis in progress

2020 — 2024
B.Sc. Computer Engineering

University of Calabria

2024 — 2025
European Patent Office

Create – Protect – Innovate: Bringing Ideas to Market

English
Cambridge B2

Certified English proficiency