CODEXIS LAB RESEARCH
Researching what comes after today's AI architectures.
Codexis Lab is interested in practical AI research: improving capability while reducing unnecessary compute, memory, latency, and deployment complexity.
Research themes
Efficient language models
Coding-focused models
Transformer optimization
Alternative architectures
Hybrid neural architectures
Retrieval and external memory
Agentic systems
Smaller models for practical hardware
Experimental research
Cortexa
Cortexa is a Codexis Lab experimental research project focused on coding-focused AI and efficient language models. Its research explores smaller models, reduced compute and memory requirements, Transformer optimization, and alternative and hybrid architectures. It is not a production-ready product.
The research question
Can coding-focused AI become substantially smaller, more efficient, and easier to run without sacrificing the capabilities developers actually need?
Research methodology
Cortexa is intended to follow an experimental approach.
Hypothesis
Define a specific architectural or system hypothesis.
Prototype
Implement a small experimental model or architecture.
Dataset
Evaluate against appropriate coding and language datasets.
Benchmark
Measure relevant dimensions such as coding capability, accuracy, memory usage, parameter count, inference latency, throughput, training cost, and hardware requirements.
Compare
Compare experimental approaches against appropriate baseline models.
Iterate
Use the results to refine the architecture and define the next experiment.
Research principle
Research should connect to measurable experiments. Hypotheses, datasets, evaluation methods, and results should be documented rather than replaced by unsupported claims. These areas are research directions, not confirmed solutions.
Research into practical AI systems.
Interested in collaborating or discussing an idea? Contact Codexis Lab.