Eve
Ask a question about the platform. Find the knowledge behind the answer.
A local AI assistant that searches QaaS documentation, docstrings and code comments to help engineers discover commands, conventions and explanations.
Local RAG implementationThe problem
As QaaS grew, its knowledge spread across documentation and source files. Engineers needed a quick way to find the right convention, understand a command or locate an explanation.
My contribution
Design and development of a local assistant within QaaS: project-content extraction, semantic retrieval, local model inference and entry points through the platform tooling.
Eve brings that knowledge into a natural-language workflow, helping people become more autonomous with the platform.
From project files to an answer.
- Extract
Read documentation, docstrings and comments. Keep source paths with the extracted passages.
- Index
Encode passages with MiniLM sentence embeddings and store vectors and metadata in FAISS.
- Retrieve
Find passages related to the question and assemble them as context for the model.
- Generate
A local llama.cpp model uses the retrieved context and a prompt requesting source references.
The Flask service, model runtime and index are packaged with the QaaS environment. Phi-2 and Mistral provide local model options. Once installed, the local runtime handles inference.
A question an engineer might ask.
Illustrative walkthrough.
How can I run API tests without installing their Python dependencies on my laptop?
Relevant project context
- The pytest launcher provides the execution entry point.
- The pytest container supplies the runtime and test dependencies.
- The selected environment determines configuration, network access and authentication.
Source paths let the engineer open the documentation or configuration behind an explanation.
The guidance to provide
Use the prepared pytest environment and the platform launcher. The container provides Python and its test dependencies; the environment configuration supplies the target and access settings.
Environment → pytest launcher → container → resultsLink the answer to the launcher documentation, required configuration and result locations so the engineer can follow the workflow.
Keep knowledge within reach.
Source-aware retrieval
Passages retain their originating files. Retrieval and source references help engineers inspect the context used for an answer and continue into the project documentation.
The knowledge base consists of extracted documentation, docstrings and comments.
Next steps
Evaluate answer quality, reference accuracy, index freshness and responses to gaps in the documentation.
A further extension would let Eve prepare and launch test campaigns through a common orchestrator.