TLDR: A new AI platform, AlteraSF, significantly reduces hiring screening time by 90-95% and detects AI-assisted or copied responses in résumés. It evaluates candidates based on factual claim validity, job fit, and linguistic authenticity, providing transparency and improving trust in AI-mediated evaluations across various job domains.
The modern hiring process is increasingly challenged by job seekers exaggerating qualifications and using AI to create polished but potentially misleading résumés. This makes it difficult for recruiters to distinguish genuine candidates from those with inflated claims, leading to more time spent manually verifying information and a decline in trust.
Altera Strategy Foundry Inc. has developed AlteraSF, an AI-native résumé-verification platform designed to address these issues. This system goes beyond traditional applicant tracking systems (ATS) by not just managing workflows but actively verifying factual claims and assessing candidate authenticity.
AlteraSF works by extracting factual statements from résumés, such as skills and achievements. It then generates unique, context-sensitive verification questions for each candidate. Responses are scored based on two quantitative measures: “Claim Validity,” which assesses the truthfulness of the claims, and “Job Fit,” which measures how well the candidate aligns with the role’s requirements.
In addition to these scores, AlteraSF version 1.1 introduced “Integrity Signals.” These are qualitative flags that detect linguistic or behavioral anomalies, such as repetitive phrasing, copy-pasting, or patterns consistent with AI-assisted writing. These signals are flagged for review by recruiters, providing an additional layer of insight into a candidate’s authenticity.
A retrospective analysis of anonymized data from pilot hiring campaigns using AlteraSF across six job families and 1,700 applications revealed significant improvements. The platform achieved a 90-95% reduction in screening time, making the process 28-150 times faster. This efficiency translates into substantial cost savings for recruiters, ranging from $50 to several thousand dollars per role depending on applicant volume.
The study also highlighted that candidate truthfulness can be evaluated not only through factual accuracy but also through patterns of linguistic authenticity. For instance, in software engineering roles, approximately 60% of “diamond” candidates (those with high Claim Validity and Job Fit) exhibited low-entropy, repetitive syntax typical of AI-assisted text, even though their factual claims were strong. This suggests that factual precision and linguistic authenticity are independent but complementary factors.
Interestingly, the rate of integrity flags varied by domain. Technical and engineering roles showed the highest frequency of AI-style regularity patterns, possibly due to the structured nature of the writing inviting generative assistance. In contrast, marketing and creative design roles, which involve more narrative and expressive writing, had lower rates of integrity flags, indicating more authentic human language variability.
AlteraSF is designed to complement existing ATS platforms like Ashby, Greenhouse, Lever, and Workday, which primarily focus on logistics and workflow management. While these systems handle scheduling and structured scorecards, AlteraSF provides the crucial layer of claim-level verification and authenticity analysis. It can function as a standalone lightweight ATS for smaller businesses or integrate as an analytics layer within larger enterprise ATS ecosystems.
This approach introduces a new level of transparency in AI-based hiring. Unlike opaque ranking systems, AlteraSF provides explainable outputs, showing the evidence for claims and the reasoning behind its evaluations. This transparency is vital for compliance with emerging algorithmic-accountability regulations and for building trust in AI-mediated evaluation systems.
Future developments for AlteraSF include expanding multilingual calibration, refining entropy thresholds for different domains to reduce false positives (where legitimate conciseness might mimic AI uniformity), and exploring temporal signature modeling to analyze response cadence. The ultimate goal is to establish empirical standards for verifiable trust in hiring.
Also Read:
- Enhancing LLM Factual Consistency with Prefix-Level Inference
- Unlocking Concise Reasoning: How Decoding Tree Sketching Improves AI Accuracy
For more details on this innovative platform, you can refer to the full research paper: Quantifying truth and authenticity in AI-assisted candidate evaluation.
