Jimoh Sunday Abdullahi

🖥️ IT Incident Resolution Workflow

The process I follow for troubleshooting, from first ticket to closed ticket — used across Tier 1–2 support and remote incident escalation.

1. 01Incident triage & logging — gather user error logs, system state screenshots, and hardware status indicators.
2. 02Root cause analysis — eliminate causes systematically: hardware → OS/driver → network → application.
3. 03Resolution & verification — apply the fix, then verify system stability post-resolution.
4. 04Documentation — log root cause and solution steps in the ticketing system and knowledge base for the team.

🎫 IT Support Case Study — Enterprise VPN & Subnet Conflict Escalation

Ticket INC-2026-8841 · Priority: High (P2 – business operations affected) · Category: Network / Remote Access / AWS VPC Routing

1. 01Initial triage — confirmed internet connectivity (ping 8.8.8.8, low latency) and verified the VPN tunnel had assigned an internal IP.
2. 02Routing isolation — ran traceroute to the internal host; traffic dropped at the default local gateway, pointing to a split-tunneling subnet conflict.
3. 03Root cause identified — ipconfig /all showed the user's home router subnet (10.200.4.0/24) matched the company's AWS staging VPC subnet exactly, causing an IP routing collision.
4. 04Resolution — set a temporary static route to prioritize VPN adapter routing, then remotely reconfigured the home router's DHCP pool to a non-conflicting 192.168.1.0/24 range and flushed the local DNS cache.
5. 05Verification — SSH connections and database ping tests succeeded, with latency under 35ms.

🤖 AI Model Evaluation Focus Areas

What I look at when grading and comparing AI model responses for RLHF and quality projects.

📋 AI Evaluation Task Walkthrough — Constraint & Factuality Audit

A real pairwise evaluation task: RLHF quality scoring on a technical/educational AI assistant, checking instruction-following, factuality, tone, and negative-constraint adherence.

Factuality check: Model A was accurate and used a clear, accessible traffic-analogy to explain router vs. switch behavior. Model B was technically accurate in networking terms but broke the evaluation protocol by using every forbidden term it was told to avoid.

🏷️ Computer Vision Annotation — QA Standards

Rules I hold annotation work to when labeling images for object detection datasets.

• Bounding box precision — boxes touch the outermost edge pixels of the object, max 1–2px margin of error.
• Occlusion under 50% — box the visible portion only, tag occluded = true.
• Occlusion over 50% — omit the box unless the project explicitly asks for full shape estimation.
• Truncation — extend the box to the image boundary, tag truncated = true.
• Low light / motion blur — expand the box to the blur halo and flag quality_flag = low_clarity for QA review.

🚗 Computer Vision Case Study — Bounding Box Quality Audit

Auditing a raw annotator batch against SOP for an autonomous-vehicle object detection dataset — sample: frame_08429.jpg, urban traffic intersection.

🐍 Dataset QA Validation Script

A Python script I wrote to catch bad records — missing fields, empty responses, or responses too short to be useful — before a dataset goes into training.

def validate_dataset_records(dataset):
    # Scans records for missing fields and length issues
    passed, errors = 0, 0

    for idx, entry in enumerate(dataset, 1):
        record_id = entry.get("id", idx)
        prompt = entry.get("prompt", "").strip()
        response = entry.get("response", "").strip()

        # Flag missing mandatory fields
        if not prompt or not response:
            errors += 1
            continue

        # Flag responses under the minimum length
        if len(response) < 10:
            errors += 1
            continue

        passed += 1

    return passed, errors