M1 · RAG
RAG Assurance
Tests retrieval quality, groundedness, and answer faithfulness. Catches hallucination, context conflation, and retrieval failure before they reach production.
M2 · LLM
LLM Evaluation
Scores model outputs for accuracy, consistency, and hallucination rate against your own ground-truth test sets — not generic third-party benchmarks.
M3 · Security
Security Testing
Probes for prompt injection, jailbreak susceptibility, data leakage, and PII exposure. Generates evidence packs ready for security and compliance review.
M4 · Agents
Agent & MAS Testing
Validates multi-step agent behaviour, tool use, and failure recovery — including multi-agent systems. Includes a published Agent Memory Benchmark.
M5 · RAI
Responsible AI
Tests bias, fairness, overreliance, and safety refusal quality. Produces audit-ready evidence packs aligned to regulatory expectations.
M6 · Continuous
Continuous Eval
Tracks model behaviour across releases, detects quality drift between deployments, and integrates into CI/CD so regressions are caught before they ship.
M7 · Multi-Modal
Multi-Modal Evaluation
Tests vision-language models, document AI pipelines, and OCR systems — covering image faithfulness, chart interpretation, and structured field extraction.
M8 · XAI
Explainability
Real SHAP and LIME attribution, counterfactual testing, and attention analysis. Quantifies how and why a model reaches its outputs — for audit and governance.
Classical ML testing is on the roadmap.
