# RecruitX AI — Complete Technical & Product Specification

## 1. Problem Statement
Modern recruitment workflows are saturated with AI-generated resumes and bot applications. When a high-growth company publishes an open role, they receive 800+ inbound applications in 72 hours. Over 85% are noisy or misaligned.
- Traditional ATS platforms (Greenhouse, Ashby, Workday) act as systems of record and avoid automatic rejection or scoring due to legal liability under AEDT regulations (NYC Local Law 144, EU AI Act).
- Human recruiters spend 20+ hours per role manually opening PDFs.
- Staffing agencies charge 15% to 25% contingency fees ($20k–$35k per placement) primarily to handle initial sourcing and screening.

RecruitX AI acts as an autonomous first-layer intelligence filter that ingests the full candidate pool and surfaces a verifiable top-tier shortlist.

## 2. Product Architecture & Two-Tier Market Model

### Tier 1: Pre-Seed & Seed Teams (Spreadsheet-Native)
- Ingestion: Tally, Google Forms, Typeform, Airtable, or direct PDF folder drop.
- Processing: Multi-signal evaluation against user-defined role rubrics.
- Output: Two-way sync with Google Sheets. Ranks candidates 1 to N with auditable scorecards and citations.

### Tier 2: Series A to C+ Teams (ATS-Native)
- Ingestion: Greenhouse Harvest Webhooks, Ashby Webhooks.
- Processing: Deterministic knockout and deep evidence verification running asynchronously via edge workers.
- Output: Pushes private scorecard notes and candidate tags (`RecruitX-Shortlist`, `RecruitX-Pass`) into the ATS. Advances top candidates directly to the recruiter interview queue while leaving the ultimate hiring determination in human hands.

## 3. Evaluation Engine Mechanics

RecruitX AI relies on a three-phase pipeline:

1. **Extraction (<20ms)**: Headless, high-speed document text extraction via PyMuPDF preserves layout structure, table layouts, and positional line metadata.
2. **Deterministic Knockout (Jev Engine)**: Fast rules evaluate non-negotiable constraints (e.g., maximum notice period, geographic/visa constraints, required certifications).
3. **Evidence Matching & Citation**:
   - Compares stated claims across form answers with verifiable tenure in resume documents.
   - Extracts exact page numbers, line ranges, and direct quotations.
   - Flags discrepancies (e.g., candidate claims 5 years experience in form field, but resume confirms only 1.5 years).
   - Generates an "Audit Receipt" rather than an unexplainable probability score.

## 4. Compliance and Legal Safety
- **No Black-Box Automated Decisions**: RecruitX AI does not issue automated employment rejections. It drafts internal evaluations and annotations for human recruiters, remaining fully compliant with AEDT guidelines.
- **Explainable Criteria**: Scorecards rely exclusively on objective criteria defined in the employer's rubric.

## 5. Technology Stack
- **Frontend**: Next.js, React, Tailwind CSS, TypeScript.
- **Backend & Compute**: FastAPI, Python (PyMuPDF), Supabase, Edge Functions.
- **Browser Client**: Manifest V3 Chrome Extension operating exclusively on active tab DOM without headless bot automation.