Our Philosophy
AI Governance & Compliance Executive Summary
Karat follows a "Human-in-the-Loop" philosophy. While AI enhances efficiency and data organization, all final hiring decisions and integrity validations are driven or verified by humans.
- Classification: Karat’s AI applications are classified as AI/ML Static AI. These models use explicitly programmed parameters and are manually trained offline; they do not autonomously retrain in production.
- Privacy & Training: No customer data (including candidate resumes or interview transcripts) is used to train the base models of our third-party AI service providers.
- Data Hosting: All AI service providers are industry leaders hosted in secure, US-based data centers.
- Review: All underlying models underwent rigorous human quality assurance and checks by Karat subject matter experts during development to ensure logic alignment.
Capability matrix
Feature Breakdown & Technical Matrix
The following table outlines the specific AI/ML models currently utilized within the Karat platform.
| Feature | Primary Function | Model Type | Data Inputs | Human Oversight |
| Resume Parsing | Extracts and formats candidate geography, education, and work history. | OpenAI GPT-4o | Candidate Resumes | Internal data review. |
| Interview Result Summaries | Synthesizes interview performance into a written summary for recruiters. | OpenAI GPT-4o | Interviewer (IVE) notes & rubrics | Opt-in feature; can be toggled off. |
| Question Timestamp Model | Identifies exact start/end times for questions in interview videos. | Sentence transformers (in-memory language models) | Transcripts* & Content Guides | High-confidence (0.75+) auto-applied; others flagged for human review. |
| Automated Integrity Review | Analyzes interview patterns to identify potential integrity concerns and flags it for review | Specialized ML Model | Candidate keystroke data | A human must review and validate or reject the concern. |
| Evidence-Based Timestamp Model | Generates timestamps for specific moments in the interview where a candidate demonstrated key skills | OpenAI GPT-4.1 | Transcripts* & Content Guides | Only available for clients using the NextGen product. Opt-in feature; can be toggled off. |
| Assessment Area Summaries | Summarizes the skills and question pools within a specific "Archetype" (job role). | OpenAI GPT-4o | Karat Content Guides | Pre-validated Karat content. |
*For these transcripts, an LLM is used to transcribe the interview
Data Privacy and Compliance
Security & Data Privacy FAQ
Is my data passed to AI providers?
If you choose to upload a candidate’s resume, that "customer data" is processed by the LLM for parsing. However, the models do not retain or train on this data.
Is any candidate data passed to the AI providers?
No. No candidate data is exposed to the LLM.
Does AI decide who we hire?
No. AI is used as a productivity aid (summarization, navigation, data formatting). Every recommendation and integrity flag requires human validation.
Can we disable AI features?
Yes. Features like Interview Result Summaries can be toggled on or off at the customer’s direction.
Karat Qualify
For customers using Karat Qualify for top-of-funnel screening, we utilize Item Response Theory (IRT).
- Some jurisdictions (e.g., NYC) classify IRT tools within the legal definition of AI/ML in employment.
- Karat provides dedicated support to ensure customers remain compliant with local AI hiring regulations when using these tools.
Last updated: May 14, 2026