Victor O.

Key Skills

Software

No software listed

Top Subject Matter

AI tutoring/knowledge assistant (multi-tenant RAG)

Content moderation / classification

EdTech tutoring / educational content generation

Top Data Types

Text

Image

Top Task Types

Question Answering

Classification

Function Calling

Freelancer Overview

AI/ML Engineer — African Institute for Artificial Intelligence (AI4AI). Brings 3+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include LangChain, HuggingFace, and Azure OpenAI. Education includes Bachelor of Technology, Federal University of Technology, Minna (2024) and Machine Learning Specialization, Stanford Online (2023). AI-training focus includes data types such as Text and labeling workflows including Question Answering, Classification, and Function Calling.

Labeling Experience

AI/ML Engineer — Algorythme

TextQuestion Answering

Designed and built an EdTech lesson-plan platform with LLM orchestration to accelerate curriculum production and improve answer grounding. Implemented a two-tier RAG pipeline that hydrates semantic chunks and anchors generated tutor responses to verified course material to reduce hallucinations. Engineered an AI profiling engine that updates student learning profiles based on chat transcript analysis to adapt tutoring pedagogy across sessions. • Built multi-tenant lesson-plan generation with FastAPI + Gemini • Implemented two-tier RAG using verified course materials as grounding • Developed transcript-driven learning profile updates in Redis • Orchestrated student-adaptive tutoring behavior across sessions

2025 - Present

AI Engineer — Mensah Philosophic Circle (security system prototype)

Text

Prototyped an AI premise security system that performs detection and supports audit-trail retrieval of evidence for suspicious content. Used retrieval mechanisms backed by vector search to find relevant prior evidence associated with detections. Work focused on improving detection accuracy and traceability for security-related decisions. • Implemented YOLOv8/InceptionResNet-based premise security prototype • Achieved 90%+ detection accuracy on premise security tasks • Added FAISS-backed audit trail retrieval for evidence • Built components to support decision traceability

2025 - 2025

AI Engineer — Mensah Philosophic Circle (Hey Nova AI)

TextFunction Calling

Co-built an Azure OpenAI educational agent that supports study workflows by generating summaries and producing question sets from course materials, with live internet access. The agent’s outputs required structuring tutoring-style content generation and retrieval of relevant information for learners. The work emphasized automated creation of learning aids rather than manual content writing. • Implemented an Azure OpenAI educational agent with live browsing • Automated summarisation for course study materials • Generated questions to support practice and assessment • Integrated agent behaviors into an end-to-end tutoring workflow

2025 - 2025

AI/ML Engineer — African Institute for Artificial Intelligence (AI4AI)

TextQuestion Answering

Deployed a multi-tenant RAG Knowledge Assistant with RBAC/ABAC access controls to prevent data leakage across tenants. Focused on retrieval-augmented generation workflows that require curating and grounding responses in verified knowledge from tenant-specific data. Implemented the system using production-grade AI orchestration and storage components for reliable AI tutoring/assistant behavior. • Built LangChain-based RAG pipelines anchored to tenant-scoped data • Configured Pinecone for retrieval and AWS S3 for knowledge artifacts • Enforced isolation via RBAC/ABAC policies and secure storage access • Supported production observability with monitoring infrastructure

2025 - 2025

Machine Learning Engineer (Intern) — Zummit Africa

TextClassification

Built a zero-shot content moderation proof-of-concept using HuggingFace BERT to classify potentially harmful or policy-violating content without relying on supervised label training. Validated the approach as a label-free alternative to supervised moderation pipelines. The work involved preparing inference inputs and evaluating model outputs against moderation needs. • Developed HuggingFace BERT-based zero-shot moderation pipeline • Tested label-free classification behavior vs supervised alternatives • Performed evaluation and validation of moderation suitability • Prototyped components needed for deployment-ready moderation

2024 - 2024

Education

Work History