ASPICE agentic QA framework

Context

Automotive SPICE assessment turns on Information Items: the documented evidence that a process was followed, cross-referenced against functional-safety obligations.

Constraint

Producing and maintaining that evidence is substantial work that competes directly with engineering, and the gaps are usually discovered during the assessment rather than before it.

Contribution

I built a multi-agent framework that generates, reviews, and manages ASPICE 4.0 Information Items with ISO 26262 cross-referencing, including a gap-analysis engine that compares what exists against what the process requires.

Approach

The system is exposed over the Model Context Protocol as tools, resources, and prompts, so it is not bound to one agent host or one model vendor and works with local or hosted backends through any compatible framework. The gap analysis runs as a staged pipeline using semantic similarity rather than exact matching, since the same requirement is rarely worded twice the same way.

Outcome

The documentation gap becomes visible while it can still be closed, and the evidence set can be maintained as the process changes.

Evidence

The work is held in a private repository containing the protocol server, the tool set, and the gap-analysis engine.