Benjamin M.

Key Skills

Software

Other

Top Subject Matter

LLM evaluation and technical annotation for cybersecurity

Cloud Domain Expertise

Identity Domain Expertise

Top Data Types

Text

Document

Image

Top Task Types

Prompt + Response Writing (SFT)

Function Calling

Classification

Question Answering

Text Summarization

Fine-tuning

Evaluation/Rating

Text Generation

Object Detection

RLHF

Segmentation

Red Teaming

Freelancer Overview

Modern Work Architect — Copilot Agents and AI output evaluation discipline (prompt/rubric/validation). Brings 15+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Other. Education includes Bachelor of Science, University of Eldoret (2013). AI-training focus includes data types such as Computer Code, Programming, and Text and labeling workflows including Evaluation, Rating, and Prompt + Response Writing (SFT).

Labeling Experience

Modern Work Architect — Copilot Agents and AI output evaluation discipline (prompt/rubric/validation)

Other

Evaluated LLM outputs for technical accuracy by flagging hallucinations, outdated guidance, and incorrect configurations against production knowledge. Used rubric-based side-by-side ranking across instruction-following, factual accuracy, code correctness, reasoning quality, and tone, producing detailed rationale writeups. Ensured evaluation outputs were validated before reaching enterprise production settings and aligned with deployment governance requirements. • Ranked model responses against explicit criteria for safety, correctness, and reasoning • Authored reference/gold responses and rubric explanations for specialized technical domains • Designed prompt boundaries and grounding approaches that affect evaluation outcomes • Performed pre-production validation of automated/AI tooling outputs

2025 - Present

Self-initiated Production Job-Hunting Agent (Python + Claude Haiku scoring)

OtherTextFunction Calling

Created an autonomous job-hunting agent that retrieves job postings and scores each candidate job using the Anthropic Claude Haiku API. Designed prompt logic to compare job descriptions to the user’s CV and produce ranked results suitable for downstream action. Implemented an evaluation pipeline in which retrieved text inputs are processed, deduplicated, scored, and compiled into daily summaries. • Pulled LinkedIn alerts via Gmail IMAP and queried Remotive/Adzuna APIs • Scored and ranked job candidates using Claude Haiku API with prompt engineering • Deduplicated records using SQLite and generated daily HTML digest output • Deployed and operated the agent via GitHub Actions free-tier infrastructure

2024 - Present

Head of Modern Workplace & Security — Sentinel KQL detections and playbook-based output validation

OtherText

Authored cybersecurity annotation-like artifacts by writing KQL detection logic and SOAR playbooks across identity, endpoint, and cloud telemetry. Validated alert quality by analyzing detection outputs for true/false-positive patterns and refining investigation logic across multiple telemetry sources. Produced structured operational documentation and analytical reasoning steps that function as guidelines for consistent labeling and triage. • Reviewed large volumes of log records to distinguish correct vs incorrect alerts • Wrote and improved KQL analytics and automated playbooks for SOC workflows • Performed endpoint, identity, and cloud investigation validation for alert classification • Authored runbooks/SOPs and documentation supporting repeatable triage decisions

2023 - 2025

Head of Modern Workplace & Security — Copilot Agent prompt/boundary and grounding artifact creation

OtherTextPrompt + Response Writing (SFT)

Authored and operationalized expert prompts and structured guidance for Copilot Agent workflows to support enterprise technical tasks. Built grounding sources, prompt boundaries, and human-in-the-loop checkpoints for bid-document analysis and change-request triage use cases. Translated practitioner expertise into training-like instruction artifacts, including reference materials and documentation used to steer consistent outputs. • Produced prompt and boundary designs for agent behavior • Integrated grounding sources and checkpoint mechanisms for reliability • Documented agent design decisions using formal architecture artifacts • Supported instruction steering through rubric-like operational constraints

2023 - 2025

Education

Work History