Three products which together cover the whole AI system lifecycle: from model development to verification & validation through to monitoring & reporting.
TIKOS™ Evaluate
TIKOS™ Evaluate is the right solution for completing AI model & system audits, assessment and assurance tasks.
After integrating with your pipeline, every time a model runs TIKOS™ collects raw metrics and aggregates into reports, designed to satisfy the requirements of dominant AI laws, regulations and standard frameworks.
- For AI audits, assessments and assurance
- Collects insights at inference time
Reports preconfigured for AI regulations
TIKOS™ Explain
TIKOS™ Explain is a first to market solution for compliant AI model transparency and explainability.
After integrating TIKOS™, internal processing and output traces from model runs feed our contextual reasoning technology. Every decision output can then be fully explained and evidenced via robust data provenance.
- Compliant explainability solution
- When interpretability alone is not enough
- Robustly explains every decision output
TIKOS™ Explore
TIKOS™ Explore combines core functions of ‘Evaluate’ and ‘Explain’ in an effective Trustworthy AI developer toolkit.
This helps developers build trustworthy AI systems by integrating reliable evaluations during development and testing phases.
- Trustworthy AI developer toolkit
- Build robust, compliant systems faster
- Includes evals for all relevant metrics
All TIKOS™ products are designed with a Regulations 1st Approach - they will help you comply with AI laws, regulations and standards by focusing on the following themes.
Fair & Unbiased
AIl AI data, models and systems must proactively mitigate bias to ensure fairness and non-discriminatory outcomes.
Clear & Explainable
AI systems and outputs must be transparent and explainable, with clear documentation of their processes and decisions.
Accurate, Safe & Robust
AI models and systems must perform reliably, resist errors and misuse, and operate without causing harm.
Accountable
AI systems must enable human oversight with clear reporting and accountability. Decision outputs should be contestable by users.
TIKOS™ is agnostic to model class, developer framework, tooling and deployment infrastructure. Importantly our products work across standard machine learning and deep learning architectures, including LLMs.
Standard ML
Many high performing systems remain powered by standard machine learning models, such as random forest and XG Boost. All TIKOS™ products can be applied to this class of model.
Deep Learning
Deep learning architectures are extremely challenging to test and evaluate. Uniquely, TIKOS™'s Synapses Logger innovation accesses and processes activation information at inference time which can be analysed to understand a wide range of model behaviours and failure classes.
LLMs & GenAI
TIKOS™ includes tooling that can be applied to both closed and open-weights models. For open-weights, such as the Llama family of models, internal activations are extracted and analysed. And for closed commercial models, proprietary input/output techniques provide a wide range of insights.
