Minimal UI. Maximum engineering.
Full‑stack developer focused on real products: AI‑augmented systems, clean architecture and fast, accessible UIs.
Federated CIFAR training with ResNet18, FedAvg, FedDF, Flower, and reporting artifacts.
Multi-agent inventory quoting and warehouse fulfillment with async agents, message protocols, and traces.
NLP CLI for dataset analysis, FinBERT evaluation, inference, optional fine-tuning, and metrics artifacts.
ECDH key exchange, derived session keys, and authenticated encryption for secure messaging.
Projects
Built a local federated-learning project around CIFAR-10 and a custom ResNet18 baseline: official dataset parsing without torchvision, centralized reference training, non-IID client splits, FedAvg, FedDF with STL-10 distillation, Flower simulation, plots/confusion matrices, and Word/PDF reporting.
Implemented two SPADE-based MAS projects: SP1 broker/inventory/quote agents for stock checking and custom-order quotes; SP2 warehouse manager, robot, and logger agents for grid-based fulfillment with reservations, path planning, typed JSON messages, and execution traces.
Built an installable Python pipeline for Financial PhraseBank sentiment classification: dataset profiling, stratified splits, pretrained FinBERT evaluation, single-sentence inference, optional fine-tuning, and reproducible metrics/confusion-matrix artifacts.
Built a secure messaging prototype with modern cryptographic primitives: ECDH key agreement, symmetric encryption, and authenticated messages. Focused on threat modeling, key derivation, and correct crypto usage (no custom primitives).
Implemented an end-to-end ML workflow: preprocessing + log-transform, K-Fold cross-validation, and evaluation with MAE/RMSE/NRMSE/R². Tuned depth and leaf size to control overfitting and ensure stable generalization.
ML engine (LSTM per‑stock) on OHLCV + sentiment. Secure Google OAuth, model versioning and explainable buy/sell rationales.
Full‑stack system for office seat management. Admin analytics, date‑range occupancy and an AI recommender (REST) using user prefs + history + constraints.
End‑to‑end emergency response app with automatic crash detection, live location sharing and one‑tap SOS alerts to contacts and services.
Autonomous driving system for a remote‑controlled car using ROS and OpenCV. Implemented lane detection, obstacle avoidance and path planning.
Streamlined marketplace for part‑time and full‑time roles aligned with academic schedules, up and running in 48 hours.
Hardware + mobile product with programmable LED patterns via BLE. Earned €40k investor commitment on national TV; multiple national hackathon/contest awards.
Mobile app concept to boost blood donations using rewards and clear guidance; 4th place at DPIT and semifinalist at Innovation Labs.
Skills & Stack
About
I’m a Computer Science & Mathematics graduate building AI‑powered, production‑grade apps. I care about clean architecture, performance budgets, testing, and developer experience.
I led teams, shipped products (mobile, web, embedded), and pitched in international competitions. I enjoy turning real constraints into elegant software.