Boma A.

Key Skills

Software

Don't disclose

Label Studio

Top Subject Matter

Content Generation

Social Media

User Profiling

Top Data Types

Text

Document

Computer Code Programming

Top Task Types

Text Generation

Classification

Entity (NER) Classification

Computer Programming/Coding

Prompt + Response Writing (SFT)

Freelancer Overview

ADEGuard: Clinical NLP Extraction & Unsupervised Clustering Framework. Professional background includes roles such as AI/ML Engineer. Core strengths include Label Studio and Groq LPU (Inference Acceleration). Education includes Bachelor of Engineering, University of Port-Harcourt (2012). AI-training focus includes data types such as Text and Document and labeling workflows including Entity (NER) Classification and Prompt + Response Writing (SFT).

Labeling Experience

Personalized LinkedIn Content Engine (Few-Shot & RAG) + LLM evaluation support

DocumentPrompt + Response Writing (SFT)

Built RAG and metadata-driven generation workflows to control output tone, topics, and retrieval precision using Llama 3 classification. Designed systems that retrieve relevant few-shot examples for improved reasoning and reduced latency in downstream generation. Established an LLM evaluation approach to measure response accuracy, hallucinations, and safety compliance. • Classified text attributes such as tone and topics to drive precision retrieval. • Generated branded LinkedIn content using few-shot prompting conditioned on historical data. • Used semantic similarity search with ChromaDB to select relevant SQL/code examples. • Evaluated model outputs for accuracy, hallucination risk, and safety compliance.

Not specified

ADEGuard: Clinical NLP Extraction & Unsupervised Clustering Framework

Label StudioTextEntity (NER) Classification

Developed a clinical NLP pipeline that transforms unstructured biomedical text logs into token-level BIO labels for NER. Built character-to-token alignment to map Label Studio highlights into BIO tags for fine-tuning a specialized BioBERT NER model. Applied weak supervision signals from patient metadata to support downstream severity prediction training. • Converted VAERS text into token-level BIO tag annotations (drug and symptom phrases). • Used a hybrid feature matrix combining sentence embeddings and demographic/modifier flags. • Derived symptom clusters via HDBSCAN to create stable labeled symptom cohorts. • Trained/updated models for automated multi-class severity prediction (Mild, Moderate, Severe).

Not specified

Education

Work History