Verifit: an Anti-Hallucination Career Engine with Sanity Agent Actions

Verifit is an AI-powered career intelligence engine, CV tailoring suite, and job application tracker designed to prevent the fabrication of false information often found in AI-generated resumes. The problem with most AI resume builders is their tendency to create fabricated years of experience, fake degrees, and inflated job titles. Employers also struggle with ATS keyword matchers that lack semantic understanding and reward buzzword stuffing over verified experience.

The unique concept behind Verifit is the creation of a zero-trust career graph. This approach treats career records and job opportunities as an auditable, structured content graph within Sanity, backed by deterministic scoring, cloud-deployed schema contracts, and human-in-the-loop workflows. In essence, Verifit turns Sanity into a verifiable AI operating system.

Key features of Verifit include multi-source ingestion and extraction, hierarchical requirement trees, zero-hallucination evidence matching, anti-hallucination CV tailoring and bespoke cover letters, and a full application pipeline. The application process consists of a custom Kanban tracking board with side-by-side before-and-after score delta analytics and an export drawer.

Verifit's interactive features include an interactive modular CV editor with dynamic section reordering, human-in-the-loop JD review stepper, grounded matches panel for real-time evaluation, and a score & requirement delta analysis panel. The application package drawer allows users to generate a one-click clipboard bundle containing their tailored CV content, cover letter, and job metadata ready for submission.

The core stack for Verifit consists of Next.js 16, React 19, Tailwind CSS v4, Lucide React, Sanity Studio v5, Sanity Agent Actions, @sanity/client API, @sanity/vision, Better Auth over Turso LibSQL & Drizzle ORM, @mozilla/readability, jsdom, ipaddr.js (SSRF pinning), mammoth, unpdf for content extraction, and various testing tools through over 200 automated unit, DOM, and Playwright integration tests.

The development process for Verifit was built through an AI-native pair programming workflow using Antigravity IDE and Claude Code. Rather than generating throwaway scripts, the focus was on prompt-driven systems architecture, enforcing modular separation of concerns, writing formal Architecture Decision Records (ADRs), and pair-prompting automated test suites to ensure models didn't introduce regressions.