Research led. Engineering driven.
We build open source tools and rigorous methods for making AI models more capable, efficient, and useful in the real world.
A model is only as useful as the measurements behind the decision to use it.
Illustrative dimensions only. No benchmark values are represented.
Model choice, training, precision, and runtime behavior should answer to the actual task.
Open source engineering
Four focused projects, one underlying belief. AI systems should be built, compared, and improved through evidence. Our work is under active development, with public availability determined by each project release.
Product 01
Model efficiency
An evidence first approach to model quantization. Designed to compare precision choices against quality, memory, hardware compatibility, and real inference constraints.
Product 02
Model specialization
Model fine tuning with the right training recipe, dataset discipline, and evaluation plan. Focused on improving task performance while tracking regressions and compute costs.
Product 03
Agentic post training
A research and engineering track for training smaller open models in tool driven environments. Its focus is reliable agent behavior, verifiable outcomes, and task success per unit of compute.
Product 04
Workflow optimization
An open source effort to inspect and improve LangGraph workflows. Built around finding optimization candidates and evaluating the impact on correctness, latency, and execution cost.
One philosophy across every project
Start with a baseline that reflects the real task.
Compare candidates under explicit conditions.
Preserve the evidence behind every decision.
Huyawo Solutions
Open source methods are the starting point. When your models, data, hardware, and business constraints are specific, Huyawo Solutions brings dedicated technical work to the problem.
Explore Huyawo SolutionsFocused technical engagements. Clear written collaboration.
Thinking in public
Our code is one part of the work. The methods, questions, and technical reasoning behind it deserve a place of their own.
01 / Technical work
Technical studies, experiments, methodology, and research notes for understanding what holds up under evaluation.
02 / Ideas and explanation
Writing about open models, efficient AI systems, engineering choices, and the questions worth examining more closely.
Built in the open
We believe useful open source and reproducible engineering should move forward together.
Find us on GitHub