Ameh Matthew Oche

The Problem

Babban Gona runs high-volume recruitment waves across multiple roles, but previously lacked structured psychometric or behavioural screening. This created a critical operational blind spot — we repeatedly encountered cases where technically skillful candidates failed post-hire due to behavioural risks or cultural misalignment. Without standardized assessments to evaluate personality and integrity, hiring decisions were highly subjective, funnel visibility was non-existent, and there was no defensible audit trail.

What I Built

A custom recruitment assessment platform that successfully introduced HEXACO personality, Personal Reaction Blank (PRB) behavioural risk, and role-specific MCQ tests into the hiring funnel. The platform features admin, recruiter, and candidate portals. Candidates self-register, sign in, and complete the assessments assigned to their active cycle. Admins create cycles, bulk-import rosters, assign assessments per candidate, manage the question bank, configure auto-flag risk rules, and review results.

Operational Visibility

A super-admin dashboard tracks pipeline volume, registration-to-assignment ratio, submission progress, completion rate, and auto-flagged candidates in real time.

The Hard Parts I Solved

• Cycle-scoped multi-tenancy. Concurrent cycles run without bleed-through. Every domain table carries a cycle_id with cascade rules, and Postgres RLS policies are the authoritative authorization boundary — not application code.
• Pre-registration assignment queue. Admins assign tests to candidates before they register. A pending-assignments table plus trigger-based reconciliation materializes assignments the moment a candidate signs up.
• Composite auto-flag rules. Server-side scoring evaluates a Honesty-Humility × PRB composite and writes flagged results into an admin review queue automatically to prevent high-risk hires from slipping through.
• Audit-ready decision trail. Every privileged action — cycle status change, candidate import, assignment, override, role change — is written to an append-only audit log via a service-role writer never exposed to the browser.
• AAL2-protected admin access. Mandatory TOTP MFA for every admin, enforced at both middleware and database layers.

Outcome

Introduced structured, data-driven behavioural screening to the organization from scratch, replacing subjective evaluations with a single auditable system. In active production use across recruitment cohorts. A recent active cycle tracked 68 uploaded candidates, 52 registered and matched, 128 distinct assignments, 92 submissions, 71.9% completion, and 8 high-risk candidates successfully auto-flagged for review before hiring decisions were made.

The Problem

Initial behavioural interviews at scale consumed hundreds of recruiter hours per wave. Manual screening lacked structure, varied across reviewers, and produced no verbatim record for downstream hiring managers.

What I Built

An end-to-end web platform featuring a structured AI video interview room. After passing a system pre-check (camera, microphone, face-visibility), candidates are welcomed by the AI agent. To ensure accessibility and clarity, the platform displays questions on-screen in real time as the agent speaks. Candidates then use a controlled click-to-speak input to record their responses. These responses are instantly processed via speech-to-text, securely saved as transcripts in the admin dashboard, and the agent seamlessly advances to the next question.

Recruiter Experience

Synchronized transcripts, audio playback, AI-generated scores, manual override controls, and proctoring flags — all in one centralized review surface.

The Hard Parts I Solved

• Ultra-low latency voice pipeline. WebRTC audio transport via LiveKit. The AI worker utilizes the OpenAI Realtime API to deliver prompts, while the controlled recording and transcription pipeline ensures flawless speech-to-text data capture directly to the admin dashboard.
• Privacy-first proctoring. In-browser computer vision via Google MediaPipe. Video frames are processed client-side, never transmitted to a server — preserving candidate privacy while still logging integrity events.
• Resilient session management. Custom state-machine handles network disconnects with a grace period, letting candidates pause and securely resume without losing progress.
• AI safety guardrails. Prompt-injection defense, out-of-scope question deflection, and a graduated escalation ladder (silent log → warning → session termination).
• Rubric-based scoring. A backend pipeline evaluates answers against specific principles and behavioural indicators, returning a scorecard with evidence excerpts cited from the candidate's own words.

Outcome

Replaced unstructured behavioural screening with a standardized, auditable platform. Currently active for candidate cohorts, significantly reducing manual screening hours while capturing perfectly structured interview data.

The Problem

Internal assessments and appraisals ran on generic third-party tools that were costly, weak on exam integrity at scale, and inconsistent in their results. Malpractice was a recurring concern, and there was no clean data handoff to HR for downstream performance and lifecycle decisions.

What I Built

A custom platform bridging admin-facing test management with employee-facing assessment delivery. Features include randomized test generation, timed exams with autosave, server-side grading, role-based access controls, and strict anti-cheat proctoring (tab-switch detection, focus loss, copy attempts, right-click blocks, and optional camera/microphone monitoring).

L&D Integration

Beyond point-in-time assessments, Excelerate expanded into the broader Learning & Development surface, enabling course authoring (chapters and modules) and learner progress dashboards to track employee upskilling and readiness.

Outcomes

• Proven at scale. Successfully deployed across two major organizational appraisal cycles in 2025 and 2026, securely assessing over 450 staff members in total.
• Flawless execution. Supported 200+ concurrent staff during high-stakes, organization-wide testing windows with zero system failures.
• Risk mitigation. Completely eliminated the malpractice risks tied to previous tooling.
• Operational efficiency. Delivered clean, instantly auditable results directly to HR, entirely removing the need for manual cross-checking and data reconciliation.

The Problem

Staff replacement requests were running through manual email chains. Requests were getting lost, analysts were being bypassed, records were inconsistent across hubs, and the process was effectively impossible to audit after the fact.

What I Built

A centralized, spreadsheet-driven approval system with Apps Script automation behind it. The tool auto-routes approvals based on hub and unit, enforces timestamping on every action, and acts as the single source of truth for all replacement requests. Every stakeholder — managers, analysts, HR — is consistently included in the flow with no manual cc'ing required.

Outcome

• 480+ replacement requests processed end-to-end with zero operational errors to date.
• Cut turnaround time on approvals by replacing async email back-and-forth with structured, deadline-aware routing.
• Made the entire process auditable — every action has an actor and a timestamp.