AI-Powered Recruitment & HR Operations Platform
A 14-workflow enterprise recruitment and people operations platform connecting candidate intake, AI-assisted rubric evaluation, commercial agreement governance, human HR approval, 29-token contract generation, 8-point compliance gating, automated onboarding, and centralized incident reliability.
Role
AI Automation Engineer & Systems Architect
Tech Stack
The Challenge: Disconnected Operations & Unstructured Handoffs
In high-volume recruitment environments, talent evaluation and hiring workflows frequently suffer from fragmented operational handoffs. Crucial data was dispersed across isolated document folders, manual spreadsheets, and unstandardized email exchanges:
Unstructured Ingestion
Resumes, job descriptions, and profile submissions arrived without uniform schema, requiring repetitive manual parsing and error-prone evaluation.
Fragmented Governance
Commercial rate cards, customer-specific terms, and agreement versioning were checked manually, risking contract discrepancies and unapproved terms.
Execution Bottlenecks
Offer letter generation, rejection dispatch, and candidate communication relied on manual copy-pasting, lacking auditability and state locking.
How to read these figures: time-to-offer compares the pre-automation backlog (14 days) with post-automation clearance (<24 hours) for that same backlog; compliance and rate-leak figures are counts from the production gates described below, not sampled estimates. Detailed evidence is shown in the private walkthrough on request under NDA.
End-to-End System Architecture
A 16-stage enterprise automation topology bridging candidate intake, AI rubric scoring, commercial rate governance, human HR review, 8-point compliance gating, and automated onboarding, featuring an explicit human authorization boundary.
16-Stage End-to-End Enterprise Automation Topology (14 Workflows · 4 Milestones)
HR business partners review AI evaluation reports, commercial margins, and candidate history before approving an offer.
Mandatory human checkpoint with cryptographic approval fingerprinting. The system cannot dispatch an offer autonomously.
{ reviewer_id, authorization_granted: true, agreed_salary, sign_off_date, action: 'APPROVE_OFFER' }Candidate Evaluation Engine
Transforms unstructured candidate submissions into structured evaluation records by orchestrating intake parsing, AI technical analysis, deterministic scoring validation, and operational persistence.
Automated Ingestion
Watches configured submission folders, downloads candidate resumes, extracts raw text content, and cleanses data before inference.
Structured AI Scoring
LLM prompts enforce strict JSON output formatting evaluating skills match, experience relevance, and role fit against the authoritative job description.
Deterministic Validation
Re-scores boundary cases, validates rate parameters, routes candidates to Shortlist, Manual Review, or Reject, and stores structured rows.
▸End-to-end evaluation workflow: document ingestion, structured text extraction, AI rubric evaluation, boundary validation, and operational data persistence.
Structured Evaluation Data
AI outputs are never left as transient model completions. Every evaluation is cast into a standardized, audited operational record with transparent confidence metrics.
Technical Match Score
Standardized percentage scoring against core technical prerequisites.
Confidence Rating
Statistical certainty score; low confidence triggers automatic human review.
Rate & Terms Verification
Deterministic validation against active commercial agreement limits.
Decision State
Explicit disposition (`Shortlist`, `Manual Review Required`, `Reject`).
▸Centralized evaluation data store displaying technical scores, confidence assessments, commercial validations, and deterministic routing states.
Configuration & Governance Layer
Eliminating hard-coded operational logic by centralizing customer rules, governance parameters, readiness gates, and system toggles into managed configuration tables.
Configurable Customer Rules
Thresholds, score gates, and role criteria are maintained externally by operators. The pipeline dynamically queries these rules at runtime.
Governance Readiness Checks
Strict pre-flight verifications inspect document template health, API connectivity, and active agreements prior to processing.
Data Separation Principle
Strict isolation between operator-maintained parameter tables and automated system-managed runtime states guarantees zero accidental overwrites.
▸Governance and configuration repository: manages customer-specific rules, readiness checks, and operational parameters without hard-coding.
Agreement Versioning & Active-State Governance
Commercial agreements and terms change over time. The platform maintains a complete version registry while establishing a single, unambiguous active-state authority pointer.
▸Agreement Version Registry: preserves version history, status changes, and superseded contract clauses for full historical compliance.
▸Active-State Authority Pointer: isolates the currently authoritative agreement, decoupling runtime execution from historical revisions.
Human HR Decision Layer
AI and automation support recruitment decisions, but final employment actions remain strictly human-controlled.
- ▸Standardized candidate dossier with resume extraction
- ▸Objective technical rubric evaluation and match percentage
- ▸Commercial compliance checks and salary benchmark guidance
- ▸Readiness flags highlighting missing information or ambiguity
- ▸Definitive hiring disposition (Offer, Manual Hold, Rejection)
- ▸Final agreed compensation package and start date parameters
- ▸Explicit approval sign-off triggering downstream execution
- ▸Override capability over any algorithmic recommendation
▸HR Operations Workspace: provides recruitment specialists with candidate summaries, scoring rationales, commercial bounds, and verified approval action triggers.
Approval Orchestration
When an HR operator triggers an action, the system validates eligibility, locks the processing state, captures an approval snapshot, and hands off to the controlled processor.
Confirms candidate is in a valid state for the requested action.
Atomically transitions state to `Processing` to block duplicate triggers.
Captures immutable snapshot of approved parameters and timestamps.
Passes verified payload to downstream controlled execution processor.
▸Approval Orchestrator: validates operator permissions, checks candidate eligibility, takes an immutable approval snapshot, and queues safe execution.
Safe Execution & Reliability
Zero-trust execution architecture protecting the recruitment pipeline against post-approval data alterations, duplicate document generation, and race conditions.
System hashes approved candidate ID, role, salary terms, and template version into an approval fingerprint.
Immediately before PDF compilation or email delivery, live row values are re-hashed.
Why this matters: If any spreadsheet cell or database record is tampered with between human sign-off and scheduled execution, the fingerprint mismatch immediately aborts delivery and notifies compliance.
▸Controlled Processor: checks already-processed guards, re-verifies the approval fingerprint, applies state locks, and routes execution safely.
Offer & Rejection Lifecycle
Approved outcomes branch into dedicated, decoupled execution lifecycles to prevent cross-contamination between contractual generation and respectful candidate communications.
- Validate terms & rate governance
- Populate Google Docs offer template
- Compile supporting schedule & PDF
- Store in secured Drive folder & log state
- Validate disposition authenticity
- Select calibrated feedback template
- Dispatch communication to candidate
- Zero offer-document generation invoked
▸Offer & Rejection Lifecycle: executes parameter validation, template merging, PDF compilation, and communication logging with strict branch separation.
Operational Monitoring & Telemetry
Centralized operational oversight divided into executive pipeline metrics and an actionable exception-handling center for HR operators.
▸Executive summary dashboard: tracks overall intake volumes, stage throughput, pipeline distribution, and system performance health.
▸Action Center: highlights cases requiring immediate operator attention, pending review items, approval holds, and boundary exceptions.
▸Full composite dashboard: aggregates pipeline distribution, technical score health, stage latencies, and real-time processing states into a single operational interface.
Engineering Decisions
Core design choices implemented to ensure absolute data consistency, ethical human governance, and fail-safe automation reliability.
Human-Controlled AI
AI models provide analytical candidate assessments, but final employment outcomes are strictly human-authorized.
Approval state gating separating AI evaluation from delivery workflows
Deterministic Rules around AI
LLM outputs are bounded by deterministic validation scripts enforcing schema, score bounds, and rate limits.
Type-checked JSON schemas and post-inference numeric sanity validations
Configuration Over Hard-Coding
Customer rules, score gates, and email templates reside in configuration stores rather than workflow code.
Dynamic runtime lookup from governed configuration tables
Agreement Version Governance
Commercial terms change; the system tracks full historical agreements while resolving a single active authority.
Historical version registry paired with an active-state pointer table
Approval Fingerprinting
Guarantees that parameters signed off by HR cannot be altered in the spreadsheet before execution.
Pre-execution hash comparison (Expected == Current) with auto-abort
Idempotency & Duplicate Guards
Network retries or repeated operator clicks must never result in duplicate offer letters or dual emails.
Row-level atomic state locks (`Processing`, `Processed`) before action invocation
Safe Communication Controls
Separates offer document delivery and candidate rejection messages into distinct execution sandboxes.
Decoupled downstream branches with dedicated payload validation
Operational Observability
Every workflow transition, evaluation score, and human approval creates permanent, queryable audit logs.
Dual telemetry capturing executive trends and actionable operator exceptions
What the System Changed
Transforming disconnected, manual administrative hurdles into a unified, governed, and transparent hiring pipeline.
Unified Recruitment Lifecycle
Connected fragmented recruitment stages—from resume intake to document generation—into one seamless, governed operational continuum.
Structured Evaluation Records
Converted raw AI model completions into structured, persistent operational data records with clear score attributes and confidence ratings.
Preserved Human Authority
Maintained strict human-in-the-loop sign-off on employment terms and candidate decisions, ensuring ethical and accountable hiring.
Decoupled Business Governance
Reduced dependency on hard-coded customer criteria by moving thresholds, score rules, and readiness checks to operator configuration tables.
Controlled Agreement Versioning
Established formal contract version tracking with an authoritative runtime active pointer, preventing commercial discrepancies.
Tamper-Proof Execution
Protected downstream processing against unapproved spreadsheet edits and duplicate actions via cryptographic approval fingerprinting.
Automated Document Generation
Replaced manual contract formatting with automated template population, PDF compilation, secure Drive organization, and delivery logging.
Centralized Operational Oversight
Created real-time visibility across pipeline stages, evaluation score distributions, bottleneck alerts, and pending action cases.
End-to-End Auditability
Established complete traceability connecting initial resume intake, AI evaluation scores, HR sign-offs, and final dispatched agreements.
Candidate Acceptance & Response Ingestion
Automates the intake of candidate offer replies directly from corporate Outlook channels, parsing multi-intent responses, resolving Google Drive candidate dossiers, and logging authoritative acceptance states.
Dedicated Ingestion Channel
Monitors candidate replies via Microsoft Graph API with subfolder isolation. Noise-suppression filters evaluated 144 non-offer backlog emails at Potential Offer Response? with zero false triggers.
Multi-Intent Response Parsing
Extracts deterministic commands (ACCEPT, DECLINE, COUNTER_OFFER) from email bodies and subject lines, matching against active Offer IDs.
Automated Dossier Ingestion
Auto-resolves candidate folder [CAND-XXXXXX] Full Name inside Google Drive [M0] Candidates Dossier and saves signed contract PDFs with SHA-256 hash tracking.
▸Central Candidate_Acceptance registry capturing Outlook response sources, reply timestamps, signed document IDs, and downstream M3 onboarding trigger states.
8-Point Compliance Verification Gate & Real-Time HR Bridge
Connects Google Sheets to n8n via an event-driven Google Apps Script webhook trigger operating with sub-second execution latency, enforcing a mandatory 100% verification gate across 8 critical employee documents before onboarding can proceed.
The 8-Point Pre-Employment Compliance Checklist:
Signed Employment Contract
Cryptographic OfferID & Doc ID match
Signed Job Application Form
Verified signature & declaration
Educational Certificates
Degree / diploma accreditation check
Results Transcript
Academic course breakdown
Past Payslip Copies
Proof of historical remuneration
NRIC / Work Pass / FIN
Legal employment eligibility verification
Passport Identification
Valid international travel document
Signed PDPA Consent Form
Data protection statutory compliance
Fail-Closed Gate Rule: Workflow 10 (Document & Onboarding Reconciler) enforces that 100% of required documents must evaluate to Verified. If even a single document is pending or flagged, downstream onboarding resource generation is physically locked.
▸Live Employee Documents compliance ledger showing granular verification statuses, Drive file IDs, source stages, and audit trails across candidates.
Automated Onboarding Engine & Enterprise Resource Delivery
Once compliance clearance is certified, Workflow 11 triggers turnkey resource provisioning—delivering cloud-linked timesheets, leave policies, claim forms, and personalized welcome materials with real-time SLA countdowns.
Turnkey Welcome Package
Renders personalized HTML welcome emails delivering Google Drive links to corporate Timesheet templates, Leave Application forms, Medical Claim forms, and HR Guides pulled from M3_Onboarding_Resources.
Start Date SLA Countdowns
Computes commencement due dates (T-7 days), progress percentages (0% to 100%), and automatically flags overdue milestones to talent operations partners.
Data Privacy Sanitization
Enforces strict data masking over sensitive candidate contact records. Personal email addresses and identity documents are protected with Gaussian vaulting and restricted access controls.
▸Granular onboarding checklist tracking delivery and completion states for corporate timesheets, leave forms, medical claim sheets, and employee user guides.
▸Centralized onboarding management dashboard calculating completion progress, start date SLAs, welcome email status, and HR escalation flags with privacy-sanitized candidate records.
Master Agreement Governance & Point-in-Time Immutability
Governs the multi-year Master Agreement rate lifecycle, enforcing executive email approval gating, scheduled effective-date activation, and absolute rate immutability for existing contracts.
1-Word Executive Email Approval Gating
Workflow 02 parses incoming Master Agreement rate sheet PDFs, stages commercial rates in MA_Update_Audit, and issues an approval request email to authorized leadership.
Effective-Date Activator & Atomic Archiving
Workflow 03 acts as the scheduled promotion engine. On midnight of an agreement's effective date, it promotes future commercial terms to active state, atomically archiving predecessor versions in Google Drive.
Production Telemetry: Point-in-Time Rate Immutability Certified
During controlled contract renewal test fixtures for candidate fixture Candidate A (Senior Systems Lead) ([FIXTURE_A_RENEWAL]), the candidate's grandfathered baseline compensation was strictly preserved and verified against the newly promoted enterprise commercial ceiling with zero data collision or retroactive overwrites. In addition, unvetted compensation exceptions tested on candidate fixture Candidate B triggered an immediate fail-closed state (BLOCKED), completely halting contract dispatch until formal exception sign-off.
Enterprise Infrastructure: Dual Admin Gateways & Zero-Trust Security
Replaces direct ad-hoc database manipulation with hardened microservice gateways, enforcing token authentication, allowlisted spreadsheet IDs, and write confirmation guards.
Sheet Admin Gateway
Exposes controlled endpoints (GET_VALUES, APPEND_ROW, UPDATE_ROW, GET_SPREADSHEET) requiring the mandatory header X-DCI-Admin-Key.
Google Drive Admin Gateway
Manages folder creation, dossier archiving, and permission management via audited webhook calls with strict directory allowlisting.
TEST-SEC-01 to TEST-SEC-04) confirmed fail-closed rejection on unauthorized tokens.Enterprise Incident Reliability & Central Error Telemetry
A 9-node Global Error Handler (Workflow 13) catches runtime exceptions across all 14 platform workflows, writing permanent audit logs and dispatching rich incident alerts with 1-click execution debug links.
Permanent Audit Trail
Every unhandled execution exception logs to the Error_Log sheet with Error ID, Workflow Name, Execution ID, Stage, and retryability flags.
Dynamic Incident Routing
Resolves alert recipients dynamically from M3_System_Settings, ensuring zero silent dropouts even during organizational configuration changes.
1-Click Triage Deep Link
Dispatches HTML incident alerts with color-coded severity badges, stack traces, and direct deep-links to the exact failed n8n execution history.
▸Centralized Error_Log data store tracking workflow execution failures, error codes, retry counts, input validation exceptions, and resolution states across all 14 production workflows.
High-Performance Data Architecture: 7 Native n8n Data Tables
Bypasses Google Sheets API quota bottlenecks and eliminates race conditions through 7 native n8n Data Tables providing in-memory atomic state locking and idempotency protection with sub-5ms internal query response times.
Client accounts, commercial tiers & margin floors
Active rate schedules, validity dates & version pointers
AI rubric scores & SHA hashes for deduplication
Row-level cryptographic approval locks
Dispatched offer hashes, Drive file IDs & timestamps
Response processing locks & document counter states
Webhook deduplication keys & replay protection
Zero rate-limit latency during high-volume spikes
Executive Video Walkthrough Showcase (4-Part Production Series)
Comprehensive, unedited screen recordings verifying the live system across all 4 production milestones. To protect client confidentiality and proprietary commercial rate cards, live video walkthroughs are available for verified engineering teams upon request.
Production UAT Walkthrough Series (4 Videos · 2.03 GB)
Part 1 - Candidate Intake, AI Screening & Evaluation (M1).mp4
Dedicated Outlook subfolder monitoring, multi-modal resume/JD parsing, OpenAI GPT-4o rubric scoring, active rate card validation, and evaluation logging.
Part 2 - HR Review, Offer Approval & Contract Generation (M2).mp4
HR review operational workspace, cryptographic approval fingerprinting, 29-token contract PDF & Schedule A compilation, and controlled transactional dispatch.
Part 3 - Candidate Response, Document Verification & Automated Onboarding (M3).mp4
Real-time Outlook reply intake, Candidate Acceptance registry, Google Drive dossier upload, Apps Script edit bridge, 8-point compliance reconciler, and welcome pack delivery.
Part 4 — Executive Operations Hub, Master Agreement Governance & Enterprise Safety.mp4
Multi-year Master Agreement rate management, 1-word email approval gating, point-in-time rate immutability, Admin Gateways (WF 12/14), Global Error Handler (WF 13), and native data tables.
Enterprise Production Outcomes
By replacing fragmented spreadsheets and manual handoffs with a 14-workflow governed platform, the organization achieved full lifecycle visibility and zero unapproved contract disclosures.
Enterprise Technology Stack
Engineered with production-grade automation tooling, multi-modal LLM APIs, and zero unnecessary dependencies.
n8n Cloud Enterprise
14 production workflows, sub-workflows & error handlers
OpenAI GPT-4o
Structured rubric evaluation & document parsing
Microsoft Outlook API
Transactional intake & response polling with subfolder isolation
Google Sheets & Apps Script
Operations workspace, audit trail & sub-second webhook edit bridge
Google Drive & Docs
Dynamic 29-token contract PDF compilation & dossier archival
7 Native Data Tables
Sub-5ms in-memory atomic locking & deduplication
Tokenized Security
X-DCI-Admin-Key authorization with fail-closed security
Global Error Handler
9-node error interceptor with responsive HTML alerts & deep links
Interested in a similar architecture for your team?
Discuss a Project