Building around the world of AI
Research across language models, reasoning, agent systems, and machine learning.
Research across language models, reasoning, agent systems, and machine learning.
Fezvo Labs is a lean independent research lab made up of part-time researchers and builders interested in language models, reasoning systems, agent workflows, and machine learning research. We focus on experimentation, open research and learning through practical work.
Current areas of focus across modern AI systems.
Studying models that can plan, coordinate actions, and complete multi-step tasks.
Exploring compact language models optimized for efficient inference.
Working on alignment, fine-tuning, and reasoning improvement methods.
Exploring structured learning and graph-based reasoning methods.
Building datasets for reasoning, evaluation, and synthetic generation tasks.
Analyzing calibration, reliability, and reasoning patterns in language models.
Research, projects, and open-source releases.
A benchmark focused on whether models arrive at correct answers through valid reasoning.
A Python framework for managing
multi-agent workflows.
View Repository
Notes, experiments, and technical articles.
A walkthrough for training Qwen models
for image-text tasks on limited hardware.
Read Article