Central administrative control surface to monitor and calibrate resume parsing, candidate scoring rubrics, semantic job matching, AI recommendation feeds, rate quotas, and statutory compliance (Ghana Act 843 & EU AI Act).

Monthly Inferences

1.43M

21.6M tokens · $1,842 compute

Semantic Matching

42.8k Vectors

142ms avg · 94.6% accuracy

Resume Parsing

18,450 CVs

1.6s latency · 99.1% parsing

Candidate Scoring

31,200 Graded

Demographic Blind · Act 843

AI Recommendations

156k Delivered

28.4% CTR · Two-Tower neural

Model Gateway Uptime

99.98%

4 Providers · 4 tenants near cap

System Overview

AI Features (6)

Semantic Matching

Resume Parsing

Candidate Scoring

AI Recommendations

Usage Limits & Quotas

Model Config (4)

Inference Logs (8)

Pipeline:

All

Parsing

Matching

Scoring

Status:

All

Operational

Warnings & Caps

Active Model Mesh & Inference Gateways

99.98% Healthy

Claude 3.5 Sonnet · Anthropic Cloud

Primary Reasoning · 1,420ms avg latency · 8.4M tokens/mo · Fallback: GPT-4o

● Healthy (0.02% err)

GPT-4o & GPT-4o-mini · OpenAI Enterprise

Rubric Scoring & Fast Tagging · 380ms avg latency · 9.8M tokens/mo · Fallback: Mistral

● Healthy (0.01% err)

text-embedding-3-large · Vector Embeddings

1536-dim vector space · 110ms avg latency · 42.8k active vectors

● Optimal

gh-ner-privacy-v2 · Accra On-Premises VPC

Biometric & Ghana Card PII Redaction · 35ms latency · Act 843 compliant

● Shield Active

Platform AI Recruitment Capabilities

Resume Parsing v2.4

Extracts KNUST, UG, Ashesi degrees, certifications, & experience. 99.1% taxonomy accuracy.

Objective Candidate Scoring

Strict JSON rubric evaluation with demographic anonymization. 0.04% calibration variance.

Semantic Vector Matching

HNSW cosine similarity indexing 42.8k candidate embeddings against job descriptions.

AI Recommendations Feed

Two-tower neural network with 12% diversity injection preventing talent discovery bubbles.

Statutory Compliance & AI Ethics Guardrails

Ghana Data Protection Act (Act 843)

Mandatory PII and Ghana Card masking prior to cross-border model inference.

Demographic Blindness Protocol

Gender, age, photos, and tribe indicators stripped from candidate scoring context.

Human-in-the-Loop Safeguard

AI scores provide decision support; final customer hiring decisions require human recruiter action.

Monthly Platform Token Quota

21.6M / 28.0M Tokens

77.1% Consumed

1 Agency Tenant (Jobberman Partner 12) reached 93.1% soft quota. Automated alert dispatched.

Managed AI Recruitment Capabilities & Pipeline Control

6 Features Configured

AI Feature & Pipeline

Assigned Model

Accuracy Metric

Avg Latency

Monthly Calls

Admin Control

Feature

Vector Matching Architecture & Index Parameters

Vector Dimensions

1,536 dimensions (OpenAI text-embedding-3-large)

Vector Similarity Distance

Cosine Distance (HNSW M=16, efSearch=64)

Minimum Match Cutoff

0.72 (Candidates below 72% hidden)

Indexed Entities

42,800 Active Resumes & Requisitions

Requisition-to-Candidate Scoring Weight Coefficients

Skills Taxonomy Overlap

40%

Exact and semantically related hard/soft skills.

Work Experience & Seniority

30%

Total years in role, responsibilities, & promotion trajectory.

Location & Work Mode

15%

Accra/Kumasi onsite, hybrid, or remote flexibility.

Salary & Comp Expectation

15%

Candidate compensation expectation vs employer budget.

Resume Extraction & African Educational Taxonomy

Tertiary Institution Mapping

KNUST, UG Legon, Ashesi, UCC, GIMPA, UPSA, Pentecost Univ

Name Normalization

Akan, Ewe, Ga, Dagbani first/day name token recognition

OCR Fallback Engine

Triggered on scanned image PDFs (<50 raw chars detected)

Structured Output Schema

JSON Schema v4 (RFC 7159) with strict validation

On-Premise PII Scrubbing Rules (Prior to Cross-Border API)

Ghana National ID / Ghana Card (GHA-XXXXXXXXX-X)

Strictly Redacted [HASHED]

Phone & WhatsApp Numbers (+233 XX XXX XXXX)

Masked (Last 4 Digits Only)

Residential Address & Postal GPS (e.g. GA-183-9201)

City/Region Preserved (Accra/Kumasi)

Objective Candidate Scoring Rubric Hyperparameters

Scoring Scale

1 to 100 Objective Index

Demographic Neutrality

Enforced (Gender/Age Blind)

Human Escalation Gate

Scores <60 or >92 trigger human review

Model Assigned

OpenAI GPT-4o (Structured Outputs)

⚡ Ghana Labour Act (Act 651) & Non-Discrimination Policy Note

All AI candidate scoring operates exclusively as an objective recommendation engine. No automated rejection occurs without authenticated recruiter oversight. Demographics are scrubbed from evaluation prompts to eliminate bias.

Two-Tower Recommender Architecture & Exploration Tuning

Diversity Injection Factor

12% Serendipity Exploration

Graduate Cold-Start Boost

+15% Visibility (First 14 Days)

Recruiter Action Weighting

Click (0.3) · Save (0.7) · Interview (1.0)

Monthly Recommendations

156,000 Delivered (28.4% CTR)

Tenant AI Usage Limits, Rate Limits & Overage Policies

4 Tiers Active

Subscription Tier & Tenants

Monthly Allocation

Tokens Consumed

Rate Limit

Overage & Throttling Policy

Action

0% used

0 RPM

Registered AI Model Providers & Endpoints

Max Tokens:

Temp:

Latency:

Cost:

Fallback:

Real-Time Inference Activity & Statutory Audit Log

8 Recent Streams

Request ID

Tenant & User

Pipeline / Feature

Model Used

Latency / Cost

Audit Status

Dossier

· t

REQ-AI-9841

200 OK · Verified

Claude 3.5 Sonnet

AI Pipeline Inference & Audit Dossier

Full telemetry, token distribution, sanitized prompt payload, and Ghana Act 843 compliance audit trail.

Telemetry & Latency

Hyperparameters

Prompt & Payload

Ghana Act 843 Safety

Tenant Context

Inference Telemetry Breakdown

INPUT TOKENS

2,420 tokens

COMPLETION TOKENS

1,420 tokens

TOTAL LATENCY

1,620ms (Gateway: 18ms)

COMPUTE CHARGE

$0.0115 (USD)

Sanitized Payload JSON

{
"request_id": "REQ-AI-9841",
"pipeline": "resume_parsing",
"status": 200,
"tenant": "Scancom MTN Ghana",
"candidate": "Kwesi Mensah",
"education": [
{"degree": "BSc Computer Science", "institution": "KNUST Kumasi", "year": 2020}
],
"experience": [
{"role": "DevOps Engineer", "company": "Hubtel Payments", "duration_months": 38}
],
"skills": ["AWS", "Docker", "Kubernetes", "Terraform", "Python", "CI/CD"],
"pii_redaction_status": "Ghana Card GHA-7291048-1 Masked [HASHED]",
"bias_demographic_parity": 0.01
}

✓ Ghana Data Protection Commission (Act 843) Audit Pass

Personal Identifiable Information (PII) including Ghana Card ID, age, and phone number were scrubbed on-premises prior to model inference. Demographic bias audit score: 0.01 (within statutory limit <0.05).