Experiments in creating a service layer to sit between UX clients and backend LLMs.

Intended to provide capabilities such as:

• prompt engineering/augumentation (adding additional context to the user supplied prompt, rejecting poor quality prompts)
• response validating/linting/filtering (stripping sensitive information, yaml validation)
• allowing selection from multiple backend models
• response attribution (match response to training data and cite the reference)
• record user feedback about response quality

$ cd cmd/wisdom $ go build .

See https://github.com/bparees/wisdom/blob/main/config/sample.cfg.yaml for an example.

$ ./wisdom serve --config path/to/config.yaml

$ ./wisdom infer --config path/to/config.yaml --prompt "write a deployment yaml for the registry.redhat.io/rhel9/redis-6:latest image with 3 replicas"