Tw0zer0s Freelance

I build compact, high-performance AI systems for real-world constraints.

From synthetic data design to model training and deployment, I help companies ship practical NLP and LLM solutions quickly.

EXPERTISE

Train and fine-tune compact LLMs for production use

Build synthetic data pipelines when labeled data is scarce

Convert unstructured documents into ontology-ready datasets

Deploy high-throughput training and inference on distributed GPUs

Design domain-specific AI solutions for regulated industries

SYNTH Generalist Dataset

Core contributor to an open synthetic dataset: 40B+ words, 80M samples, 8 European languages.

Synthetic DataMultilingualReasoning Models

GPU MODE Hackathon

Trained a small language models (400M parameters) on multi-GPU multi-node infrastructure.

PretrainingModel EfficiencyHugging Face

RATP Distress Detection

Built an end-to-end pipeline with 1.7M synthetic tweets and a 600M model matching larger proprietary systems.

Synthetic DataFine-TuningPublished to ACL 2026

Bacardi Global Forecasting

Improved sales forecasting error by 30%+ and deployed the system across products and markets.

Time SeriesSupply ChainProduction ML

Let's Build Something

Available for freelance projects in LLM training, synthetic data, and production ML systems.