9 projects
aevyra-forge
Autonomous overnight optimizer for LLM inference deployments — tune vLLM config, quantization, and kernels jointly against your real workload.
aevyra-origin
Failure attribution for agent pipelines — given an AgentTrace and a score, Origin finds which node(s) caused the failure.
aevyra-witness
AgentTrace — the record of what happened during an agent run. The shared trace primitive for the Aevyra stack (Reflex, Origin, Verdict).
aevyra-reflex
Agentic prompt optimization — an AI agent that diagnoses eval failures and rewrites prompts until your model hits the target score
aevyra-verdict
Benchmark any LLM against your data. Pick the best model, then make it better.
torch-workflow-archiver-nightly
Torch Workflow Archiver is used for creating archives of workflow designed using trained neural net models that can be consumed by TorchServe inference
torchserve-ag
TorchServe is a tool for serving neural net models for inference
torch-workflow-archiver-ag
Torch Workflow Archiver is used for creating archives of workflow designed using trained neural net models that can be consumed by TorchServe inference
torch-model-archiver-ag
Torch Model Archiver is used for creating archives of trained neural net models that can be consumed by TorchServe inference
