Evidence-First Governance

Every decision,
audited.

Resume screening that stands up to scrutiny. No inferred scores. Just what matched, why it matched, and who approved it.

Full Audit Trail
Evidence-Based
Defensible

Pilot · Internal testing only · Not for live hiring decisions

How It Works

Evidence-First, Every Step

From resume upload to final approval, every decision is captured, reviewed, and defended with full transparency.

Deterministic Evidence

Evidence is extracted, not inferred. Every match is backed by direct quotes from the resume with clear justification.

Multi-Model Adversarial Review

A second AI pass challenges the first. Disagreements are surfaced. Consensus is required before human approval.

Human Approval Gate

A recruiter cannot approve their own rubric. A second reviewer must sign off. Decisions are never unilateral.

For Recruiters

Defend Every Screening Decision

Upload job rubrics with criteria, thresholds, and required keywords. TalentGuard scans resumes and produces an audit artifact for every decision.

Upload job rubrics tied to specific job descriptions
Every screening decision produces a tamper-evident audit artifact
Full rubric version history — every edit tracked and preserved
Role separation — each user sees only their namespace

For Candidates

See Exactly How You Stack Up

Upload and version your resume. Import job listings from any portal. Get real-time keyword suggestions and a freshness check against the current job market.

Import job listings via URL — LinkedIn, Naukri, any portal
Compare your resume against multiple jobs simultaneously
Freshness check — flags outdated tech and stale phrases
Real-time AI keyword suggestions for your target role

Built for Trust

Governance by Design

Every action is logged. Every decision is tracked. Every user has clear, enforced boundaries.

SQLite Audit Log

Every action is time-stamped and logged. No action is untracked. No decision is hidden from the audit trail.

Role Separation

Admin, recruiter, candidate — each role sees only their namespace. No data leakage. No cross-role access.

OTP-Based Login

No passwords. One-time passcodes only. Simpler, more secure, and fully auditable authentication.

Shadow Evaluation Mode

Results cannot be written to an ATS or used to rank or reject. Evaluation only — no integration with live hiring systems.

What It Is NOT

By design, TalentGuard is intentionally limited. These constraints protect both recruiters and candidates.

Not a Ranking Engine

TalentGuard does not rank candidates or recommend who to hire. You make the decisions — it gives you the evidence.

Not ATS-Integrated

No writes to Greenhouse, Lever, or any ATS. Results stay in TalentGuard only.

Not Bias-Audited

Not legally cleared for production hiring. Use as a governance tool alongside human judgment, not as a replacement.

Not Connected to Job Boards

No scraping, no OAuth to LinkedIn or Naukri. You import jobs manually via URL. You stay in control.

Evidence-First Screening Starts Here

Join recruiters and candidates building fairer, more transparent hiring processes.

Screening Screen a Resume
?
SHADOW EVALUATION ONLY — prototype output; not approved for hiring decisions, candidate ranking, ATS updates, or candidate communication.

Drag and drop resume PDF / DOCX here, or click to browse

Supported: PDF, DOCX, TXT, MD, RTF, HTML, ODT

or paste text directly
Governance Result

Smart Scan

Gemini evaluates resume evidence directly against a specific job description.

Semantic Evaluation

Job Match

Gemini ranks the resume against the configured professional job tracks.

Career Fit Recommendations
Evidence and scores use only content from your uploaded resume. Candidate-added evidence is labelled pending verification and does not increase the source-resume score. No application is submitted without your explicit confirmation.
① Profile & Resume
Loading…
② Select Portal
③ Import Job
④ Compare & Score
Total Cases
Review Needed
Evidence Aligned
Evidence Gaps
Highlighted rows need comparison feedback. Feedback does not affect any candidate or hiring process.
Case Evidence Finding Policy Check Comparison Status Date
Loading cases…

Model Governance

Runtime model inventory, assigned roles, controls, and declared limitations.

Loading model governance…

Rules Administration

Inspect risk controls and create versioned rubric drafts without modifying approved rules.

Rubric Requirements

Select any rubric to inspect it. Saving creates a new unapproved draft; the current rubric remains unchanged.

MUST-HAVE REQUIREMENTS
PREFERRED REQUIREMENTS

Draft Review and Approval

Compare the selected draft with its parent rubric. Approval requires passing activation validation and a different authenticated reviewer from the draft creator.

Select a draft to review its controls.

Risk Flag Catalog

Loading risk controls…

Settings

Governance configuration and privacy controls for this instance.

Privacy Controls
Resume text stored
Source filename stored
Retention period
API binding127.0.0.1 (local only)
Rubric Library
Total approved
DISCO verified
Demo prototypes
Blocked (contaminated)
Governance Status
Activation validatorrubric_activation_validator_v1
Scannerdeterministic_resume_scanner_v1
Policy engineresume_policy_v1
AuthenticationPrototype session — email identity, in-memory tokens
Anonymised Decision Records

When a candidate's personal data is erased, a PII-free decision record is kept here. HR can use these to respond to right-to-explanation or challenge requests under EU AI Act, UK GDPR, or NYC LL144.

Loading…
Candidate Challenges

When a candidate challenges a hiring decision, log it here. The system links it to their anonymised record and generates a formal response letter.

Log a New Challenge
The reference is created with the governance case and should appear in the candidate notice. Loading available references…
Loading challenges…

Create recruiter invite

Limited to approved company domains. Send the link only through the recruiter's company-controlled chat. Anyone holding the link and knowing the assigned email can use it before expiry.

Recent invites

Active invites can be revoked. Used, expired, and revoked links cannot create another session.

Loading…
Evidence Agreement Rate
Assessments Recorded
Disagreements
Permitted Pilot Use
• Compare evidence quality against an existing recruiter tool
• Use synthetic, de-identified, or formally approved historical resumes
• Review whether requirement matches are supported by exact resume evidence
• Record recruiter feedback about false matches, missed evidence, and usability
Prohibited Use
• Do not use prototype output as the basis for advancing, rejecting, ranking, shortlisting, or contacting a candidate
• Do not upload live applicant data unless you have documented company approval and are specifically authorized to process that data in this prototype
• Do not write or transfer prototype results into Greenhouse or any other applicant tracking system
• Do not represent this build as legally compliant, independently bias-audited, validated for employment decisions, or production-ready
Known Technical Gaps
• Keyword and model matches can be incomplete, incorrect, or overly literal
• Scanned-image PDFs are not supported without OCR
• Candidate notice, state-transition, identity, and artifact-integrity controls are not production-grade
• Synthetic bias testing does not establish fairness on real applicant populations
• Applicable laws and obligations require current review by qualified company counsel
Pilot Success Measure

The pilot tests one narrow claim: whether showing exact resume evidence for every requirement gives recruiters a more useful and reviewable comparison than their current tool. Positive recruiter feedback does not prove legal compliance, predictive validity, or absence of discrimination.

Case Detail

Scan Summary

Human Approval Gate

Loading required action…

Candidate Notice

Generate a compliance notice to send to the candidate informing them AI was used to screen their application.

Privacy Controls

Deletes this governance case and its linked scan artifact. A PII-free anonymised record and hashed deletion receipt are retained for audit and challenge response.