Olayiwola Ibrahim

Key Skills

Software

AWS SageMaker

CVAT

Google Cloud Vertex AI

Labelbox

Label Studio

Prodigy

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Computer Code Programming

Document

Image

Text

Top Label Types

Audio Recording

Classification

Computer Programming Coding

Fine Tuning

Question Answering

RLHF

Transcription

Freelancer Overview

I am an AI engineer and Data Science undergraduate (GPA 4.8/5.0) at University of Lagos with hands-on experience preparing, structuring, and evaluating high-quality training data for AI systems. As the founding AI engineer at PilotOps, I built a multi-tenant RAG-based agent platform where I designed semantic chunking pipelines, implemented metadata-aware retrieval, and performed response validation against ground-truth sources. This required dataset cleaning, structured annotation, hallucination detection, and human-in-the-loop evaluation to ensure reliable, source-backed outputs — directly aligning with AI training and labeling workflows. Beyond system design, I have worked extensively with structured and unstructured data, including normalization, tagging, classification logic, and analytics tracking across platforms like Quotients Africa and Jobifiy Inc. My technical strengths in Python, SQL, JSON structuring, and data validation enable me to deliver precise, consistent, and scalable training datasets. What sets me apart is my deep understanding of how labeled data impacts model performance — I don’t just annotate data; I design and evaluate it with the end model quality in mind.

Labeling Experience

Speech Data Annotation & Whisper Fine-Tuning (Nigerian Accent Adaptation)

Aws SagemakerAudioFine Tuning

This project focused on improving automatic speech recognition (ASR) performance for Nigerian-accented English by preparing and validating high-quality speech datasets for fine-tuning a Whisper-based transcription model. Responsibilities included curating and cleaning audio samples containing diverse Nigerian accents, performing accurate manual transcription, and aligning transcripts with corresponding audio segments. I normalized text outputs to maintain consistent spelling standards while preserving accent-specific linguistic characteristics. Special attention was given to code-switching patterns (English mixed with local expressions), pronunciation variations, and phonetic inconsistencies common in regional speech.

This project focused on improving automatic speech recognition (ASR) performance for Nigerian-accented English by preparing and validating high-quality speech datasets for fine-tuning a Whisper-based transcription model. Responsibilities included curating and cleaning audio samples containing diverse Nigerian accents, performing accurate manual transcription, and aligning transcripts with corresponding audio segments. I normalized text outputs to maintain consistent spelling standards while preserving accent-specific linguistic characteristics. Special attention was given to code-switching patterns (English mixed with local expressions), pronunciation variations, and phonetic inconsistencies common in regional speech.

2025 - 2025

LLM Response Evaluation & Annotation

Aws SagemakerTextQuestion Answering

This project focused on evaluating and annotating AI-generated responses to improve large language model (LLM) performance within a retrieval-augmented generation (RAG) system. The primary objective was to enhance output reliability, factual accuracy, and contextual alignment through structured human evaluation and feedback generation. Responsibilities included annotating model outputs against verified ground-truth documentation, identifying hallucinations, detecting incomplete or misleading answers, and flagging logical inconsistencies. Responses were categorized based on relevance, factual correctness, contextual alignment, clarity, and source attribution compliance. I developed structured grading rubrics to ensure consistent evaluation standards and created labeled feedback datasets to refine retrieval quality and reduce model hallucination rates.

This project focused on evaluating and annotating AI-generated responses to improve large language model (LLM) performance within a retrieval-augmented generation (RAG) system. The primary objective was to enhance output reliability, factual accuracy, and contextual alignment through structured human evaluation and feedback generation. Responsibilities included annotating model outputs against verified ground-truth documentation, identifying hallucinations, detecting incomplete or misleading answers, and flagging logical inconsistencies. Responses were categorized based on relevance, factual correctness, contextual alignment, clarity, and source attribution compliance. I developed structured grading rubrics to ensure consistent evaluation standards and created labeled feedback datasets to refine retrieval quality and reduce model hallucination rates.

2025 - 2025

Education

Work History