Recruiting Agent App Overview

Automate candidate screening with AI-powered resume analysis, generate personalized outreach emails, coordinate interview scheduling, produce offer letters, and rank applicants — all driven by a configurable recruiting AI agent built into MangoApps Workforce.

---

What is Recruiting Agent?

Recruiting Agent is the AI-powered recruiting automation engine inside MangoApps Workforce. It uses GPT-4o to parse resumes (PDF, DOCX, images), score candidates against job requirements across four weighted dimensions (technical fit, experience, education, and cultural fit), and take autonomous follow-up actions — from sending acknowledgment emails to scheduling phone screens. Every AI decision is logged in an immutable audit trail, and a configurable approval workflow keeps humans in the loop for high-stakes actions like offer letters and rejections.

The agent integrates with the Job Board (job postings and candidate applications), Interview Scheduler (interview coordination), Offer Manager (offer letter generation), and the Automation Hub (autonomous action configuration and trust-level management).

Core Value Proposition:

• 🤖 AI Candidate Screening — Parse resumes from uploaded files (PDF, DOCX, images via OCR) and score candidates 0–100 against job requirements using configurable weighted criteria
• 📧 Automated Outreach — Generate personalized acknowledgment, rejection, phone screen invite, interview invite, and passive outreach emails with AI
• 📅 Interview Coordination — Schedule, reschedule, cancel, and complete interviews with calendar integration and bulk scheduling
• 📝 Offer Generation — Produce AI-drafted offer letters with compensation summaries, saved as drafts for approval
• 🏆 Candidate Ranking — Rank all applicants for a job by AI score with configurable weight factors

---

At a Glance

🤖 AI Screening	📧 Outreach	📅 Interviews	📝 Offers
Resume parsing, 4-dimension scoring, recommendations	Acknowledgment, rejection, phone screen, passive outreach	Schedule, reschedule, cancel, calendar view	AI-generated offer letters with compensation data

🏆 Ranking	✅ Approvals	📊 Analytics	🔧 Playground
Score-based candidate ranking per job	Pending/approved/declined workflow with bulk actions	Screenings by day, score distribution, outreach stats	Admin tool testing with live and test-mode execution

Perfect For:

• 🏢 HR / Recruiters — Screen high volumes of applicants automatically, send timely outreach, and focus review time on borderline candidates
• 👔 Hiring Managers — Review AI screening results, approve or override recommendations, and track interview progress
• ⚙️ Admins — Configure scoring weights, approval rules, autonomous behaviors, and test AI tools in the Playground

---

How It Works

Recruiting Agent Lifecycle

┌─────────────────────────────────────────────────────────────────────────────┐
│                     RECRUITING AGENT LIFECYCLE                              │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  ┌───────────┐    ┌───────────┐    ┌───────────┐    ┌──────────────┐       │
│  │ CANDIDATE │───▶│  RESUME   │───▶│    AI     │───▶│ RECOMMENDATION│       │
│  │  APPLIES  │    │ EXTRACTED │    │ SCREENING │    │   GENERATED   │       │
│  └───────────┘    └───────────┘    └───────────┘    └──────┬───────┘       │
│                                                             │               │
│                    ┌────────────────────┬──────────────────┐│               │
│                    ▼                    ▼                  ▼▼               │
│            ┌──────────────┐   ┌──────────────┐   ┌──────────────┐         │
│            │   ADVANCE    │   │    REVIEW    │   │   DECLINE    │         │
│            │  (Score ≥70) │   │ (Score 30-69)│   │  (Score <30) │         │
│            └──────┬───────┘   └──────┬───────┘   └──────┬───────┘         │
│                   │                  │                   │                  │
│                   ▼                  ▼                   ▼                  │
│          ┌──────────────┐   ┌──────────────┐   ┌──────────────┐           │
│          │ PHONE SCREEN │   │   HUMAN      │   │  REJECTION   │           │
│          │   INVITE     │   │   REVIEW     │   │    EMAIL     │           │
│          └──────┬───────┘   └──────┬───────┘   └──────────────┘           │
│                 │                  │                                        │
│                 ▼                  ▼                                        │
│          ┌──────────────┐   ┌──────────────┐                              │
│          │  INTERVIEW   │   │   APPROVE /  │                              │
│          │  SCHEDULED   │   │   OVERRIDE   │                              │
│          └──────┬───────┘   └──────────────┘                              │
│                 │                                                           │
│                 ▼                                                           │
│          ┌──────────────┐                                                  │
│          │    OFFER     │                                                  │
│          │  GENERATED   │                                                  │
│          └──────────────┘                                                  │
│                                                                             │
└─────────────────────────────────────────────────────────────────────────────┘

---

🤖 AI Candidate Screening

The core engine of Recruiting Agent. When a candidate applies (or an admin triggers screening), the system extracts resume content, parses it into structured data, scores it against job requirements, and produces a recommendation — all in a single optimized AI call.

Resume Extraction Pipeline

The service supports multiple file formats with a prioritized extraction strategy:

1. PDF files — Local extraction via pdf-reader gem first; falls back to OpenAI Files API for scanned/image-based PDFs
2. DOCX files — Local extraction via docx gem first; falls back to OpenAI for complex layouts
3. DOC files — OpenAI Files API extraction (legacy Word format)
4. Image files — OpenAI Vision OCR for resume screenshots or scans
5. Structured data fallback — If no file is attached, the system builds a resume from candidate profile fields (name, email, experience, education, skills, certifications)

Four-Dimension Scoring

Every screening produces scores across four configurable dimensions:

Dimension	Default Weight	What It Measures
Technical Fit	40%	Skills match against required and preferred skills listed in the job posting
Experience Match	30%	Relevance and duration of work experience relative to role requirements
Education	20%	Degree level and field alignment with job requirements
Cultural Fit	10%	Cover letter tone, career trajectory, values alignment with company culture

Each dimension receives a score from 0–100. The overall score is a weighted sum using configurable weights that must total 100.

Screening Results

Each completed screening record includes:

• Overall Score (0–100) — Weighted composite
• Dimension Scores — Technical, experience, education, cultural fit
• Confidence Level — 75% (mid-range scores), 85% (near-threshold), or 95% (clearly strong/weak)
• Recommendation — advance, review, or decline based on configurable thresholds
• Reasoning — 2–3 sentence AI-generated explanation
• Strengths — 3–5 specific strengths from the candidate’s background
• Gaps — 2–3 areas where the candidate doesn’t fully match
• Deal Breakers — Critical missing requirements (if any)

Screening Actions

Action	Description
Screen Candidate	Trigger AI screening for a single candidate (runs as a background job)
Bulk Screen	Screen multiple candidates at once — each queued as a separate background job
Re-Screen	Force fresh resume extraction and complete re-analysis, clearing cached data
Approve	Manually override to advance a candidate, regardless of AI recommendation
Reject	Manually override to decline a candidate, regardless of AI recommendation

Screening Filters

The screenings index supports filtering by:

• Recommendation type (advance / review / decline)
• Score range (min and max)
• Job posting
• Date range
• Candidate name or email search

---

📧 Automated Outreach

Recruiting Agent generates personalized candidate emails using GPT-4o, with support for five outreach types:

Outreach Type	Purpose	Auto-Send Support
Acknowledgment	Confirm receipt of application	✅
Rejection	Respectful decline with encouragement	✅
Phone Screen Invite	Invite qualified candidates to a phone screen	✅
Interview Invite	Invite to on-site or video interview	❌ (manual)
Passive Outreach	Cold outreach to passive candidates highlighting fit	❌ (manual)

Outreach Workflow

┌─────────────────────────────────────────────────────────────────┐
│                    OUTREACH WORKFLOW                             │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  ┌──────────┐    ┌───────────┐    ┌──────────┐    ┌──────────┐ │
│  │ GENERATE │───▶│   DRAFT   │───▶│  REVIEW  │───▶│   SEND   │ │
│  │  (AI)    │    │  CREATED  │    │  & EDIT  │    │  EMAIL   │ │
│  └──────────┘    └───────────┘    └──────────┘    └──────────┘ │
│       │                                                │        │
│       │          ┌───────────┐                         │        │
│       └─────────▶│ AUTO-SEND │─────────────────────────┘        │
│    (if enabled)  │ (skip     │                                  │
│                  │  review)  │                                  │
│                  └───────────┘                                  │
│                                                                  │
│  Optional:  ┌────────────┐    ┌──────────┐                     │
│             │  AI REVIEW │───▶│ QUALITY  │                     │
│             │  (checks)  │    │  REPORT  │                     │
│             └────────────┘    └──────────┘                     │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Outreach Features

• AI-Generated Content — Subject line and body personalized with candidate name, position, and company context
• Draft Editing — Edit subject and body before sending; only drafts can be modified
• AI Quality Review — Run an AI review against quality criteria before sending (available per-outreach)
• Bulk Generation — Generate outreach for multiple candidates at once with a single action
• Send / Retry / Cancel — Send immediately, retry failed messages, or cancel unsent drafts
• Delivery Tracking — Track status (draft, sent, failed, cancelled) with sent timestamp

Outreach Stats

The outreach index dashboard shows:

• Total outreach messages
• Sent count
• Draft count
• Failed count
• Sent today

---

📅 Interview Coordination

Manage the full interview lifecycle — from scheduling through completion — with support for multiple interview types and bulk operations.

Interview Types

• Phone Screen — Initial screening call
• Video Interview — Remote video interview
• On-Site Interview — In-person interview
• Technical Assessment — Skills-based evaluation
• Panel Interview — Multi-interviewer session

Interview Actions

Action	Description
Schedule	Create a new interview with candidate, interviewer, type, and time slot
Reschedule	Move to a new time with reason tracking (stored in feedback history)
Cancel	Cancel with required reason
Complete	Mark as completed with outcome (advance / decline / hold) and notes
Bulk Schedule	Schedule interviews for multiple candidates at once

Calendar View

A dedicated calendar view displays interviews grouped by date within a configurable date range, making it easy to see daily interview load and identify scheduling conflicts.

Interview Stats

• Total interviews
• Currently scheduled
• Completed
• Cancelled
• Upcoming (future scheduled)
• Today’s interviews

---

✅ Approval Workflow

High-stakes recruiting actions require human approval before execution. The approval system provides a structured review process with audit trails.

Actions Requiring Approval

By default, the following actions require approval (configurable per business):

• Sending outreach emails
• Scheduling interviews
• Generating offer letters
• Updating candidate status (optional)

Approval Process

1. The AI agent or autonomous executor creates an Action record
2. An Approval request is generated with priority, summary, and affected entities
3. Admins and managers review pending approvals
4. On approval, the action is executed automatically via ActionExecutorService
5. On decline, the requester is notified with the decline reason

Approval Features

• Pending / Approved / Declined tabs with pagination
• Bulk Approve — Approve multiple requests at once
• Related Approvals — View other approvals for the same candidate
• Permission Checks — Requesters cannot approve their own requests; requires manager or admin role
• Timeout Handling — Approvals have a 48-hour timeout window
• Notification Integration — Requesters are notified on approval or decline

---

🏆 Candidate Ranking

Rank all candidates for a specific job posting by their AI screening scores, with configurable weight factors applied.

How Ranking Works

1. Select a job posting
2. The system screens any candidates who haven’t been scored yet (using the same screening service)
3. All candidates are ranked by their weighted overall AI score (highest first)
4. Results show candidate name, email, score, and the weight factors applied

Ranking Configuration

Weights are applied from the business configuration:

┌───────────────────────────────────────────────────┐
│              SCORING WEIGHTS                       │
├───────────────────────────────────────────────────┤
│                                                    │
│   Technical Fit:  ████████████████████  40%        │
│   Experience:     ██████████████       30%         │
│   Education:      ████████             20%         │
│   Other Factors:  ████                 10%         │
│                                                    │
│   Total:          ████████████████████████████ 100%│
│                                                    │
└───────────────────────────────────────────────────┘

Results can be limited to a specific number of top candidates or show the full list.

---

📝 Offer Generation

Generate AI-powered offer letters for qualified candidates using the Offer Generation Tool.

Offer Workflow

1. Select a candidate (typically from the “interviewed” or “ready_for_offer” status)
2. Provide position title, base salary, and start date
3. The AI generates a complete offer letter with compensation summary
4. The offer is saved as a draft requiring approval (configurable)
5. Once approved, the offer can be sent to the candidate

Offer Data

Each generated offer includes:

• Offer Letter — Full AI-generated offer letter text
• Compensation Summary — Structured breakdown of salary and benefits
• Candidate Details — Name, position, start date
• Approval Status — Draft, pending approval, approved, or sent

---

🔧 Playground (Admin)

An admin-only testing environment where administrators can test each AI tool individually against real or test data without affecting production workflows.

Available Playground Tools

Tool	What It Tests
Candidate Screening	Test screening with an existing candidate or custom resume data (JSON or file upload)
Outreach Generation	Generate outreach emails for existing candidates with selectable outreach type
Interview Scheduling	Test scheduling with candidate and interviewer selection, mock time slots
Offer Generation	Generate offer letters for interview-stage candidates with salary and start date
Candidate Ranking	Rank candidates for a selected job with configurable result limits

Each tool supports two modes:

• Existing Data — Select a real candidate and run the AI tool against their actual profile
• Test Mode — Provide custom data for testing without creating permanent records

Test-mode screening creates a temporary candidate record, runs the full screening pipeline, and cleans up the temporary data on completion.

---

🤖 Autonomous Mode

When enabled, Recruiting Agent can operate autonomously — detecting opportunities and taking actions without manual triggering. The autonomous system runs on a scheduled interval (every 15 minutes) and includes safety controls.

Autonomous Capabilities

Capability	Setting	What It Does
Auto Screening	auto_screening_enabled	Automatically screen new applicants when they apply
Auto Acknowledgments	auto_send_acknowledgments	Send acknowledgment emails to new applicants within 24 hours
Auto Phone Screen Invites	auto_send_phone_screen_invites	Invite candidates scoring ≥ 75 to phone screens
Auto Rejections	auto_send_rejections	Send rejection emails to candidates scoring < 40

Safety Controls

• Agent Safety Service — Validates safety before each execution
• Progressive Trust Levels — Agents earn greater autonomy through successful actions; trust level determines which actions can bypass approval
• Daily Action Limit — Configurable cap on autonomous actions per day (default: 50)
• Confidence Thresholds — Only high-confidence actions (≥ 95%) bypass approval for acknowledgments
• Approval Fallback — Phone screen invites and rejections always require approval regardless of trust level
• Immutable Audit Logs — Every autonomous action is logged and cannot be modified after creation

Opportunity Detection

The Opportunity Detector scans for three types of opportunities:

1. Unacknowledged Applications — Candidates with “applied” status created in the last 24 hours
2. High-Scoring Candidates — Screened candidates with scores ≥ 75 in “screening” status
3. Low-Scoring Candidates — Screened candidates with scores < 40 in “screening” status (last 7 days only)

---

📊 Analytics Dashboard

Track recruiting agent performance with comprehensive analytics across configurable date ranges.

Available Metrics

Metric	Description
Screenings by Day	Daily screening volume chart
Score Distribution	Breakdown across 5 score bands (0–30, 30–50, 50–70, 70–90, 90–100)
Recommendation Breakdown	Count of advance, review, and decline recommendations
Outreach Performance	Messages sent and breakdown by outreach type
Interview Metrics	Scheduled, completed, and cancelled interview counts
Time to Screen	Average time from application to screening completion
Average Score	Mean screening score across all candidates

Date Range Options

• Today
• This week
• This month
• This quarter
• Last 30 days (default)

---

📱 Mobile Support

Recruiting Agent includes a full mobile web experience accessible via the MangoApps native app. The mobile interface provides:

• Dashboard — Stats overview, recent screenings, pending approvals, and recent activity
• Screenings List — Browse and filter screenings by recommendation type with infinite scroll
• Screening Detail — View full screening results including scores, strengths, gaps, and reasoning
• Approvals List — View pending, approved, and declined approvals with status filtering
• Approval Actions — Approve or decline requests directly from mobile with notes/reason

---

⚙️ Settings & Configuration

General Settings

Setting	Description	Default
Agent Enabled	Master on/off switch for the recruiting agent	Off
Autonomous Enabled	Enable autonomous opportunity detection and execution	Off
Auto Screening	Automatically screen candidates when they apply	Off
Auto Send Acknowledgments	Auto-send application acknowledgment emails	Off
Auto Send Rejections	Auto-send rejection emails to low-scoring candidates	Off
Auto Send Phone Screen Invites	Auto-send phone screen invites to high-scoring candidates	Off
Auto Schedule Interviews	Enable automatic interview scheduling	Off

Screening Configuration

Setting	Description	Default
Technical Weight	Weight for technical skills scoring	40
Experience Weight	Weight for experience match scoring	30
Education Weight	Weight for education scoring	20
Other Factors Weight	Weight for cultural fit and other factors	10
Min Qualification Score	Minimum score to avoid automatic decline	30
Auto Advance Threshold	Score at or above which candidates auto-advance	70

Content Settings

Setting	Description
Company Culture Values	Text description of company culture for cultural fit scoring
Screening Instructions	Custom instructions for the AI screener (auto-reject criteria, prioritization rules)

Sample Data

Admins can load or clear sample recruiting data (jobs, candidates, interviews) for testing purposes directly from the Settings page.

---

🔐 Bias Protection & PII Safety

Recruiting Agent includes built-in safeguards aligned with responsible AI principles:

Bias Guard

• Pre-Screening Validation — Before every screening, the system validates that scoring factors are based only on legitimate job-related criteria (skills, experience, education, certifications)
• Bias Warnings — Any detected bias risk is logged to the audit trail for transparency
• Unbiased Explanations — AI-generated explanations focus exclusively on job-relevant qualifications
• Fair Scoring Guidelines — The AI is instructed to avoid discrimination and focus on hard skills and soft indicators of success

PII Protection

• Input Masking — Candidate PII (names, emails, phone numbers) is masked before being sent to the LLM in conversational contexts
• Secure File Handling — Uploaded resumes are downloaded to temp files, processed, and immediately cleaned up
• OpenAI File Cleanup — Files uploaded to OpenAI’s Files API for extraction are deleted after processing

Audit Trail

• Every screening, outreach, interview, and approval action is logged to an immutable audit trail
• Audit logs capture: action type, status, autonomous flag, confidence score, affected entities, timestamp, and executing user
• Logs cannot be updated after creation (enforced by model callback)
• Summary reports can be generated for any date range

---

🔗 Integration Points

Integration	How It Connects
Job Board	Candidates apply through job postings; screenings reference job requirements
Interview Scheduler	Interview scheduling redirects to the Interview Scheduler app for full calendar integration
Offer Manager	AI-generated offers are created via the Offer Generation Tool and managed in Offer Manager
Automation Hub	Autonomous mode configuration, trust levels, and agent safety managed via Automation Hub
Ask AI	The Recruiting Agent is available as a conversational tool through the Ask AI interface
Notifications	Approval requests, approval outcomes, and outreach sends trigger in-app notifications
Email (Mailer)	Outreach emails are delivered via RecruitingNotificationsMailer

---

Navigation Tabs

Tab	Icon	Visibility	Description
Dashboard	fa-th-large	All users	Stats overview, recent screenings, pending approvals, recent activity
Screenings	fa-search	All users	Browse, filter, and review AI screening results
Approvals	fa-check-circle	All users	Pending, approved, and declined approval requests
Outreach	fa-envelope	All users	Manage outreach emails with status tracking
Interviews	fa-calendar-alt	All users	Interview list, calendar view, scheduling actions
Analytics	fa-chart-line	All users	Performance metrics and charts
Playground	fa-flask	Admin only	Test AI tools with live or mock data
Settings	fa-cog	Admin only	Configure agent behavior, weights, and autonomous settings

---

Frequently Asked Questions

Q: How accurate is the AI screening?
The screening uses GPT-4o with structured prompts and configurable scoring weights. Confidence levels are reported with each result (75–95%), and the system recommends human review for mid-range scores (40–69). All screenings can be manually overridden.

Q: Can I customize the scoring weights?
Yes. Admins can adjust the weights for technical fit, experience, education, and other factors. Weights must sum to 100. Changes apply to all future screenings.

Q: What file formats are supported for resume upload?
PDF, DOCX, DOC, and common image formats (JPG, PNG). PDFs and DOCX files are extracted locally first for speed, with OpenAI as a fallback for scanned documents.

Q: Will the agent send emails without my approval?
Only if you explicitly enable auto-send settings. By default, all outreach is saved as a draft for human review. Even in autonomous mode, phone screen invites and rejections always require approval.

Q: How does the autonomous mode work?
When enabled, the agent checks for opportunities every 15 minutes — new unacknowledged applications, high-scoring candidates needing phone screens, and low-scoring candidates for rejection. Each action goes through safety checks and progressive trust levels before execution.

Q: Is candidate data safe?
Yes. PII is masked in conversational AI contexts, uploaded files are processed in temporary storage and cleaned up immediately, and all actions are logged to an immutable audit trail. Bias guards validate that screening decisions use only job-relevant criteria.

Q: Can I test the AI tools before going live?
Yes. The Playground (admin only) lets you test screening, outreach, interview scheduling, offer generation, and ranking with either real candidates or custom test data — without affecting production workflows.