Roy K.

Key Skills

Software

Google Cloud Vertex AI

Top Subject Matter

Informal economy

civic grievances

administrative data processing

Top Data Types

Text

Top Task Types

RLHF

Classification

Freelancer Overview

AI Automation Developer & Systems Architect — Model Tuning & Evaluation for Kigo OS (KangwanaTech / Kigo AI). Brings 15+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Internal, Proprietary Tooling, and Google Cloud Vertex AI. Education includes Bachelor of Arts, University of Nairobi and Kenya Certificate of Secondary Education, Miathene Secondary School. AI-training focus includes data types such as Text and labeling workflows including RLHF, Classification, and Evaluation.

Labeling Experience

AI Automation Developer & Systems Architect - KangwanaTech

TextRLHF

Built and operated an AI automation system that structures unstructured WhatsApp inputs into organized administrative data using LLM tuning and evaluation practices. Designed agent logic to categorize complex civic grievances into actionable outputs with strong grounding and fact-checking workflows. Managed backend reliability and cloud deployment on DigitalOcean to deliver localized AI services with high uptime. • Developed “Kigo OS” for headless LLM-based categorization and structuring • Conducted manual RLHF-style refinement to reduce hallucinations in logging • Integrated Google Search API grounding for truth-anchored sports and news telemetry • Managed infrastructure deployment and monitoring to sustain 99.9% uptime

2025 - Present

AI Automation Developer & Systems Architect — Grounding & fact-checking for AI agent outputs (KangwanaTech / Kigo AI)

Google Cloud Vertex AIText

Integrated live search grounding using Google Search API to verify factuality of AI agent responses. Evaluated output accuracy by grounding responses in real-time external information for sports and news telemetry. Reduced response errors by ensuring claims align with retrieved evidence. • Evidence-based verification for AI-generated text • Real-time grounding for fact-checking during telemetry responses • Output quality improvements driven by evaluation against sources • Safety and accuracy alignment for technical/narrative content

2025 - Present

AI Automation Developer & Systems Architect — Systemic logic for 'Kigo Ombudsman' (KangwanaTech / Kigo AI)

TextClassification

Designed an autonomous categorization agent that converts complex civic grievances into actionable administrative data labels. Structured unstructured WhatsApp inputs into organized categories for downstream processing. Implemented logic that supports consistent label assignment for recurring grievance types. • Categorization label design for civic grievance types • Agent logic for multi-step interpretation and label selection • Structured output formatting for database ingestion • Quality improvements via iterative testing of categorization accuracy

2025 - Present

AI Automation Developer & Systems Architect — Model Tuning & Evaluation for Kigo OS (KangwanaTech / Kigo AI)

TextRLHF

Performed manual RLHF to reduce model hallucinations in data logging for unstructured WhatsApp inputs. Evaluated and tuned LLM behavior to improve structured categorization outputs. Iterated on model responses to achieve higher-fidelity recordings aligned to user-intended categories. • LLM-based categorization of WhatsApp/SMS content • Human review loop for labeling quality and hallucination mitigation • Optimization of agent outputs for structured administrative records • Integration with downstream data logging systems

2025 - Present

Education

Work History