John H.

Key Skills

Software

Other

Don't disclose

Top Subject Matter

AI agent evaluation

model behavior research

and AI governance evidence systems

Top Data Types

Document

Computer Code Programming

Text

Top Task Types

Fine-tuning

Freelancer Overview

Founder & Systems Builder — Awakened Intelligence (AI evaluation and systems builder). Brings 33+ years of professional experience across legal operations, contract review, compliance, and structured analysis. Core strengths include Other and Don't disclose. Education includes Coursework, Orange Coast College (1992) and Coursework, University of Nevada, Reno (1994). AI-training focus includes data types such as Computer Code, Programming, and Document and labeling workflows including Evaluation, Rating, and Fine-tuning.

Labeling Experience

Compliance Aegis / AIValuations — AI Governance & Evidence Systems

OtherDocument

Designed and evaluated dual-control AI compliance architectures where answer generation and evaluation were separated to enable independent verification. Built evidence-oriented workflows for regulated AI use cases, producing structured evidence artifacts suitable for governance and operational review. Collaborated on legal/technical framing to translate system behavior into Article 25-style deployed-system evidence language. • Developed dual-control evaluation flows so evaluators had different objectives than generators. • Implemented evidence framing for governance contexts including finance, healthcare, and employment. • Collaborated with an EU AI Act attorney on deployed-system evidence and AIValuations work. • Produced regulatory-aligned evidence language derived from technical model behavior.

2024 - Present

AI Interpretability Research — “What Did Your LoRA Actually Learn?” (research pipeline and evaluation)

Don't discloseDocumentFine-tuning

Contributed to interpretability research work assessing what LoRA fine-tuning changes in internal model representations using experimental design and analytic pipelines. Used multiple evaluation and ablation-style techniques to test hypotheses about representation changes that may not be visible from surface behavior alone. Supported compute, pipeline architecture, and technical collaboration for manuscript development. • Examined internal representation shifts induced by LoRA fine-tuning. • Implemented analyses using sparse autoencoder/crosscoder-style methods and cross-reconstruction ratios. • Applied causal ablation and holdout validation with scale comparisons across model families and corpora. • Supported NeurIPS submission of the manuscript through experimental and technical collaboration.

2024 - Present

Founder & Systems Builder — Awakened Intelligence (AI evaluation and systems builder)

Other

Designed and evaluated AI agent and compliance systems where model outputs were assessed using traces, benchmarks, and verifier-based completion checks. Focused on turning observed behaviors into structured evaluation signals and iterate-ready findings for improved reliability and safety. Led architecture and test strategy work to diagnose stalls versus genuine errors and reduce counterproductive interventions. • Built Grok terminal-coding harness evaluation runs (scored trials) with reviewer interventions grounded in execution traces. • Performed failure analysis to separate ordinary errors from true stalls and improve completion behavior. • Established benchmark-aligned objectives emphasizing verified task completion over internal scoring artifacts. • Produced evidence-backed technical documentation for evaluation and deployment readiness.

2024 - Present

Education

Work History