Research & Projects

Activity

1. 2026

   Poster

   TAROS 2026: Towards Autonomous Robotic Systems

2. 2026

   Track Editor

   ARC - AI Workflows for Applied Research and Evaluation International Workshop 2026 ↗

   Faculty of Economics and Business Administration, Alexandru Ioan Cuza University of Iași

3. 2025

   Inventor

   Agentically-Orchestrated Foundational Models for Cloud Architecture Development ↗

   Google · US20250343728

4. 2024

   Reviewer

   ICDCC-2024: International Conference on Data, Computation and Communication

Patents

Agentically-Orchestrated Foundational Models for Cloud Architecture Development

US20250343728 · Application ID 18651871

1. User response information is obtained, comprising information indicative of cloud architecture requirements for a cloud architecture to fulfil.
2. A plurality of agentic orchestration models - each a machine-learned language model prompted to fulfil a corresponding cloud-architecting role - generate role outputs, one of which proposes generic component placeholders for components necessary to meet the requirements.
3. Based on the role outputs, a proposed architecture output is generated, comprising a visual representation of the proposed generic component placeholders.

Access patent ↗

Projects

nanoMMO - Building an MMO

A YouTube series building an MMORPG from scratch, client and server, with the focus on the engineering around the game: a raylib and ImGui client, a C++ server kept in sync over protobuf and Asio, accounts in MySQL with Argon2id, Redis login tokens, Prometheus and Grafana, profiling with perf and Tracy, and Docker and CI throughout. Like Karpathy's nanochat, the code is kept minimal and easy to hack on, and tagged at each step of each episode.

C++ Networking Observability Video series

Website ↗ YouTube ↗ GitHub ↗

mvGPT - Minimum Viable GPT

A language model from scratch, to understand one from the ground up: a GPT defined in PyTorch and taken through the whole training pipeline - pretraining, fine-tuning and reinforcement learning - with the evaluation and metrics that show what it learned, then deployed so anyone can talk to it in the browser. Built episode by episode on YouTube, with the models on Hugging Face.

PyTorch LLMs RL Video series

Website ↗ YouTube ↗ GitHub ↗ Hugging Face ↗

InteligentArtificiala.ro

Artificial intelligence documented in Romanian, from the ground up: most technical AI material is written in English, so this builds a Romanian reference with the same rigour. A general-to-specific path runs from basic concepts through machine learning, neural networks and deep learning to how LLMs like ChatGPT, Gemini and Claude are built - tokenisation, embeddings, attention, the Transformer, pretraining, fine-tuning and alignment - and run, with inference, RAG, tool use, agents and evaluation. Each concept starts from intuition before formulas, with runnable code and the English term alongside the Romanian one.

AI LLMs Education Romanian

Website ↗ GitHub ↗ YouTube ↗

smollerGraph - Building a Graph Database

A YouTube series building a small graph database from scratch in C++, from the bytes on disk to the query language: how nodes and relationships are laid out in pages on disk, how the database follows them from one to the next, a buffer pool and indexes, and how a Cypher query becomes a plan that runs, measured along the way against Neo4j and Kùzu. It is heading towards a knowledge graph that mvGPT can ask for facts, with a browser UI that draws query results with D3. Every step of every episode is a tag in the repository.

C++ Databases Graphs Video series

Website ↗ YouTube ↗ GitHub ↗

M.I.N.E.R.V.A. Research Platform

An invite-only research workspace for academics: a library that builds itself by automatically extracting papers, side-by-side PDF and note viewing, AI assistants for summarising and drafting, citation-graph visualisation, and customisable workspaces that adapt to how researchers actually work.

Research AI Platform

Visit Minerva ↗ GitHub ↗ YouTube ↗

ML-Enabled Game Engine (Hades)

An exploration of game-engine architecture: cross-platform Vulkan, WebGL, ImGui, JoltPhysics (used in Horizon Forbidden West), SoLoud, ECS, and C++ scripting, with particular attention to creating APIs for ML training.

C++ Vulkan ECS ML

GitHub ↗

GradientHound

An adaptation of BloodHound from cybersecurity, brought to model architectures and policies - extracting insights across checkpoints into the state of the gradients with architecture-aware visuals.

ML Visualization

GitHub ↗

LaTeX++

A C++ parser for math expressions that looks up each glyph's outline in a loaded OpenType font (TTF / OTF) and emits an extruded triangle mesh ready for any renderer. No graphics-API dependency: the library returns flat vertex/index buffers for custom rendering.

C++ Rendering OpenType

GitHub ↗

Vanilla MuJoCo - JoltGym

Exploring educationally whether using Jolt for multi-agent training can be as (or more) efficient than MuJoCo - recreating the physics of environments like Humanoid / HalfCheetah, making Gym wrappers, and training RL.

RL Physics

GitHub ↗

Hades Neural Engine - C++ RL Framework

Leveraging TorchLib and custom integration layers with ImGui and Wandb to enable re-use of RL methods in other C++ projects - particularly to build and train agents that can act in 3D environments alongside human-controlled agents.

C++ RL TorchLib

GitHub ↗